Address round-1 review feedback on fetchTracking redaction:
- Cross-reference REDACTED_VALUE / REDACTED_HEADER_VALUE so a reader
understands why two identical "[REDACTED]" constants coexist.
- Document that redactBody deliberately scopes to form-encoded + JSON
(multipart/binary bodies pass through; OAuth never uses multipart).
- Add a tracker-path test asserting a nested sensitive key in a JSON
request body is redacted in the recorded copy while the live outbound
body stays byte-identical.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3fTN8H3R8YV4yUGvZjYnX
Extends the header redaction from #1585. Before a fetch entry reaches any
sink, mask known-sensitive credentials in the request/response bodies and
the URL query string, in addition to headers:
- redactBody(): form-urlencoded + JSON bodies, recursive over nested
objects/arrays, best-effort (never throws), values-only.
- redactUrlQuery(): masks sensitive query params, preserves path,
non-sensitive params, and any fragment.
- Sensitive fields: client_secret, code, refresh_token, access_token,
id_token, code_verifier, client_assertion, assertion, password, token
(case-insensitive), reusing the [REDACTED] sentinel.
Applied to the request success + error entries and the async response-body
callback. Live outbound requests are byte-identical — only the logged copy
is redacted.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3fTN8H3R8YV4yUGvZjYnX
Round-3 @claude review of #1592 (optional nit): getClientRegistrationKind is
the other synchronous getter that, like getScope, is safe only because its
sole caller (buildOAuthConnectionState) awaits getClientInformation/
getServerMetadata first, flushing hydration before this runs on the
post-redirect callback path. Add a docstring mirroring getScope's so a future
refactor doesn't read it without a preceding awaited storage read.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3fTN8H3R8YV4yUGvZjYnX
Round-2 @claude review of #1592 caught a seventh member of the clear
family that the first pass missed: clearClientInformation still wrote to
the store synchronously, carrying the same pre-hydration clobber risk the
clearAfterHydration helper was written for (a clear before hydration lands
persists a near-empty blob, overwriting the whole on-disk oauth.json).
Wrap its mutation the same way and add a clear-before-hydration test
proving a sibling token (and thus every other server's blob) survives.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3fTN8H3R8YV4yUGvZjYnX
Address round-1 @claude review of #1592:
- OAuthStorageBase.clear* no longer mutate the pre-hydration (empty) store —
a clear issued before hydration is deferred until it lands, so it merges
onto the real persisted state instead of (a) being resurrected by the late
rehydrate merge or (b) persisting a near-empty blob that clobbers every
other server's on-disk credential. Already-hydrated clears stay synchronous.
- store.setServerState migrates a pre-normalization raw-key blob onto the
canonical key on first partial write (was: shadowed it with a fresh
canonical entry, orphaning the raw blob's other fields). clearServerState
now also drops the raw-key orphan.
- Document the keepalive 64KB combined-body ceiling on the remote-storage
write, the getRemoteOAuthStorage memo key intentionally omitting fetchFn,
and the real ordering invariant behind the sync getScope.
- Tests for the clear-before-hydration and raw-key-migration paths.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3fTN8H3R8YV4yUGvZjYnX
undici 8.5.0 requires Node >= 22.19.0 (its declared engines), so CI's
Node 20.x could not load undici's CacheStorage (missing
webidl.util.markAsUncloneable) and the proxy integration tests failed.
Bump the CI setup-node to 22.x to match the repo's Node floor.
Reconcile the declared floor with undici's real requirement: bump root
engines.node to >=22.19.0 and correct the CLI README note (undici 8.5.0
needs >=22.19.0, not >=22.7.5). Also document proxy support in the web
README and clarify that a caller-supplied eventSourceInit.fetch bypasses
the proxy wrapper by design.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3fTN8H3R8YV4yUGvZjYnX
Address round-1 review nits on apps.ts:
- readUiMeta now accepts `unknown` and narrows internally, removing the
three `as WithUiMeta` casts at its call sites (readability win).
- Add a comment documenting that the content-block-then-result precedence
for `_meta.ui` is intentional (posture lives on the content block; no
shallow merge).
- Add a TODO to switch to the named McpUiToolMeta/McpUiResourceMeta types
once the upstream extensionless re-export is fixed for NodeNext.
Coverage: apps.ts 100% lines / 97.05% branch / 100% funcs — clears the gate.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3fTN8H3R8YV4yUGvZjYnX
Migrate the web client's OAuth storage from BrowserOAuthStorage
(sessionStorage) to RemoteOAuthStorage, which POSTs through the backend's
/api/storage/oauth route to ~/.mcp-inspector/storage/oauth.json (mode 0600).
A credential obtained in the browser is now the same blob the TUI/CLI read
on the same host, giving web ⇄ TUI ⇄ CLI parity (Wave 1 of #1579).
Core (shared with TUI/CLI):
- OAuthStorageBase now drives a single async hydration (`ready()`,
`getHydrationError()`); every post-redirect read (getClientInformation,
getTokens, getCodeVerifier, getServerMetadata, getIdpSession) and every
save awaits it so a late hydration merge cannot clobber a fresh write.
getCodeVerifier/getServerMetadata become async; the store is created with
skipHydration:true so there is no auto-hydration to race the explicit one.
- store.ts adds normalizeServerUrl so a token saved under
https://Example.com/mcp is found when the CLI asks for
https://example.com/mcp/; getServerState falls back to the raw key.
- remote-storage adapter POSTs with keepalive so the write survives the
OAuth authorize redirect, surfaces swallowed persist failures, and gives
richer read/write error messages.
- generateOAuthState throws instead of silently degrading to Math.random.
- NodeOAuthStorage honors MCP_INSPECTOR_OAUTH_STATE_PATH for isolated
fixtures. Provider codeVerifier()/getServerMetadata() + the EMA idpOidc and
connection-state readers await the now-async storage.
Web:
- environmentFactory + a new shared lib/remoteOAuthStorage accessor
(memoized per {baseUrl, authToken}) wire RemoteOAuthStorage into the
connection path, the EMA IdP hook, and the per-server "clear OAuth" action
so all three share one in-memory view of oauth.json.
Tests: async-hydration + normalizeServerUrl suite, keepalive/error-path
coverage, and the sync→async migration across existing storage/provider/EMA
tests. Web validate + coverage gate + integration suite green; CLI/TUI/
launcher validate green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3fTN8H3R8YV4yUGvZjYnX
Wrap the Node transport's fetch with an undici EnvHttpProxyAgent dispatcher
when a standard proxy env var is set, so the CLI and web backend can reach
remote MCP servers through corporate proxies with zero new flags. undici is
imported lazily (only when a proxy var is present) and added as a dependency
at the repo root and in the CLI client; a missing undici raises an actionable
error. Documents the behavior and dependency in the CLI README.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3fTN8H3R8YV4yUGvZjYnX
generateOAuthState() now throws when crypto.getRandomValues is unavailable
instead of silently degrading to a non-cryptographic Math.random() fallback —
OAuth state is a CSRF token and must be unpredictable.
The web /oauth/callback handler now rejects a returned `state` param that does
not parse to the expected 64-char-hex authId shape (a forgery indicator),
surfacing a clear error toast instead of proceeding with an undefined session.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3fTN8H3R8YV4yUGvZjYnX
Add a machine-readable summary of an MCP App's UI metadata: the AppInfo
interface plus extractAppInfo(), which merges a tool's _meta.ui
(resourceUri, visibility) with the linked UI resource's _meta.ui (csp,
permissions, domain, prefersBorder, mimeType). Content-block matching is
exact-URI first, then a lowercase + trailing-slash normalized match, then
a single-block fallback since resources/read returns the requested
resource by definition. Shared plumbing for the programmatic-review path
(CLI --app-info, integration tests) without rendering the app.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3fTN8H3R8YV4yUGvZjYnX
Redact Authorization, Cookie, Set-Cookie, Proxy-Authorization, X-Api-Key,
and x-mcp-remote-auth to [REDACTED] in createFetchTracker before any
request/response entry reaches the in-memory log, pino logger, or
persisted session storage. Comparison is case-insensitive; original key
casing is preserved and the live outbound request is untouched.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3fTN8H3R8YV4yUGvZjYnX
* refactor(web): update OAuth E2E tests to use new authentication method
- Replaced `runGuidedAuth` with `authenticate` in the InspectorClient OAuth end-to-end tests to support quick authentication with CIMD pre-registration.
- Updated comments to reflect the change in authentication method and its support for HTTP test metadata URLs.
- Ensured that the tests continue to validate the expected authorization URL after the change.
Additionally, imported `ensureCimdClientRegistration` in the OAuthManager to streamline client registration during the authentication process.
* Guided-auth removal, OAuth smoke testing, and TUI/CLI support
* Added CLI OAuth runner flags tests
* Added clarifying comment on port 0 binding
* README notes on TUI oauth callback port, prettier fixes for CLI tests
* CIMD validation uses inline like EMA IdP
* Fixed TUI test failures from merge
* test: bring 5 OAuth files over the 90 per-file coverage gate
The guided-auth-removal work added/touched several files that fell below
the repo's >=90 per-file coverage gate (npm run coverage). CI doesn't run
the gate (it's local-only), so validate + test:integration passed while
these gaps slipped through. Bring them all back over the line:
- core/auth/node/runner-oauth-callback.ts: test bad-URL/non-http-scheme
throws, no-port->80 default, and IPv6 hostname bracketing. Annotate the
provably-dead guards (empty host, out-of-range port, String(err)
fallback, pathname "/" fallback) -- all rejected by new URL() upstream --
with justified v8 ignore comments instead of lowering the gate.
- core/auth/cimd.ts: test the AS-doesn't-support-CIMD early return.
- core/react/useClientSettingsDraft.ts: test the functional-updater path,
the stale-onChange-after-close prev===null guard, and the
timer-fires-after-close value===null guard.
- ClientSettingsForm.tsx: test the CIMD checkbox toggle and CIMD URL clear.
- clients/tui/src/utils/oauthDisplay.ts: test static/dcr registration
kinds, the expired IdP session, and the configured-scope/empty-scope
branches of formatScopes.
npm run coverage now passes (exit 0) across web, cli, tui, launcher.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: cliffhall <cliff@futurescale.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(web): defer IdP issuer URL error to onBlur in Client Settings (#1544)
The Enterprise-Managed Authorization Issuer field validated its URL inline on
every render, so the 'Must be an http(s) URL' error nagged while the user was
still mid-typing (e.g. before 'https:/…' is complete).
Gate the issuer error display on an issuerTouched flag set onBlur; once
touched it tracks the value live, so the error clears as soon as a valid URL
is entered. The persist gate (canPersistClientSettingsDraft → validateClientSettings)
is unchanged and still validates independently, so an invalid issuer is never
written to the backend regardless of touched state.
Closes#1544
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(web): reveal issuer error and block close on invalid issuer (#1544)
Extend the on-blur deferral so an invalid issuer can't be silently dropped on
modal close (the auto-save flush would otherwise discard it with no feedback).
ClientSettingsModal now intercepts its close (X / Esc / overlay): if the issuer
is invalid it reveals the field error (via the form's revealIssuerError prop,
overriding on-blur gating) and keeps the modal open instead of closing. The
user can fix the URL or clear the field — an empty issuer is valid to leave.
The block lives in the modal (unit-tested) rather than App, so App.tsx stays a
thin flush+close that only runs when the close is allowed; the persist gate is
unchanged. Reveal state resets via the modal's open/close remount key.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(core): require a real host in isAbsoluteHttpUrl, not just an http(s) scheme (#1544)
The issuer validation used URL.canParse + an http(s) protocol check, which the
WHATWG parser accepts for bare/degenerate values that aren't actually usable
URLs — e.g. https://foo, https://example, https://. , https://.. all passed.
Tighten the shared rule (used by both the Zod config schema and the form's
inline validation): after parsing and the http(s) scheme check, require the
host to actually look like a host — localhost, an IPv6 literal, or a dotted
domain/IPv4 with at least two non-empty labels. Single-label, bare-dot, and
empty-label hosts are now rejected. Add core tests for the accepted
(dotted/IPv4/IPv6/localhost) and newly-rejected cases.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The Enterprise-Managed Authorization "Issuer" field accepted arbitrary
non-URL text and only failed at save time, surfacing a raw Zod issue array
in "Failed to save client settings" toasts.
- Inline field-level validation: the Issuer input shows a friendly error
("Must be an http(s) URL, like https://idp.example.com").
- canPersistClientSettingsDraft gates on issuer validity (deferring to
validateClientSettings) so an invalid value is never sent to the backend.
- Extract isAbsoluteHttpUrl in core/client/config-parse so the Zod schema
and the form share one http(s)-only URL rule.
- Harden the fallback save-error toast via formatClientConfigLoadError.
Closes#1533
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(web): audit branch-coverage outliers, raise per-file gate to 70 (#1271)
The web per-file coverage gate had drifted red: the recently-landed EMA
auth feature (core/auth/ema/*) and ClientSettings components shipped
undertested, and the prior 50% branch floor let business-logic branches
go untested. CI does not run the coverage gate, so this accumulated.
Add tests to bring every failing file above the gate and push branch
coverage well past 70 across the board, then raise the per-file branches
threshold 50 -> 70 in clients/web/vite.config.ts.
- EMA core (emaFlow, resourceContext, jwt, wire): 50-88% -> 86-100% branch
- EMA OIDC/transport (idpOidc, tokenEndpoint, transportProvider): -> 97-100%
- oauthManager / discovery: 73-83% -> 88-92%
- ClientSettingsModal/Form, ConnectionInfoContent: 0-88% -> 96-100%
- useClientSettingsDraft / useEmaIdpLoginState hooks: 20% -> 100%
- TaskStatusBadge / SortableServerCard outliers: 50% -> 100%
Coverable branches reached by mocking useComputedColorScheme (dark theme),
dnd-kit useSortable (isDragging), and fetch/OIDC paths. Remaining sub-100
branches are genuinely-unreachable defensive guards or happy-dom-inherent
(portal / useMediaQuery / typeof window) and documented in test comments.
No source files changed. Global web branch coverage 88.73% -> 90.63%.
AGENTS.md updated; cli/tui/launcher stay at 50 (out of scope).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test: raise per-file coverage gate to 90/90/90/90 across all clients
Follows the #1271 web branch-coverage audit (50 -> 70) by raising the
whole per-file gate to >=90 on all four dimensions — lines, statements,
functions, branches — for web, cli, tui, and launcher (core/ is measured
under the web suite). A codebase-wide audit added real tests for every
outlier and brought global coverage up across the board.
Config: all four vitest/vite coverage configs set to 90/90/90/90.
Two pure-type modules (core/mcp/import/types.ts, clients/web/server/
types.ts) added to the web coverage exclude list, matching the existing
type-only exclusions.
Tests: ~50 test files extended/created across core (auth, ema, mcp,
remote, state, storage, react), web components/screens/server, and the
cli/tui clients. Highlights: full EMA OAuth flow + transport coverage,
inspectorClient/remote server backfill, every Mantine control's handler
branches, and the cli unsupported-method / argv-fallback paths.
Genuinely-unreachable branches are annotated at the source with justified
`/* v8 ignore -- <reason> */` comments rather than lowering the gate:
happy-dom-inherent paths (Mantine portal, useMediaQuery, typeof window
SSR guards), React StrictMode effect replay, and provably-dead defensive
guards (?? fallbacks the types make impossible, Select.onChange values
outside the allowed list). No runtime logic changed; the only structural
edits are two behavior-preserving refactors (ResourceTemplatePanel hoists
the field value to a local so the ignore applies in JSX; ToolsScreen
swaps a ternary for an if) and prettier line-wrapping.
Also fixed a flaky real-timer debounce assertion in PromptArgumentsForm
(asserted an exact sibling-typing interleaving that raced the 300ms
debounce; now asserts the load-bearing fact — a non-empty fire-time
sibling capture).
AGENTS.md updated to describe the uniform 90 gate and the v8-ignore
policy. All four client coverage gates, root validate, storybook (370),
and smoke pass.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(web): make connect-crash SSE reader stop on the terminal event
`readSseEvents` looped until the server closed the stream (`done`). The
server closes the events stream immediately after writing the terminal
`transport_error`, but under the heavy parallel scheduler pressure of a
full `coverage` run (v8 instrumentation across all workers) that close
can lag far enough behind the event to trip the 30s test timeout — a
hang, not a wrong result. This made the test intermittently fail under
`npm run coverage` (it passed in isolation and under CI's lighter
`test:integration`).
Stop reading as soon as the terminal `transport_error` event arrives
(cancelling the reader) instead of waiting only for `done`. That event is
always the last one the session queue emits, so nothing is lost, and the
test no longer depends on stream-close timing. The `done` break is kept
for any stream that closes first.
Verified green across two consecutive full `npm run test:coverage` runs
(3094 tests) where it previously hung, plus isolation (~0.5s).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds enterprise-managed authorization (EMA): browser-side IdP login + RFC 8693 token-exchange and JWT-bearer resource-token mint, per-server EMA opt-in in Server Settings, IdP config in Client Settings, and IdP clientSecret keychain storage. Includes the v2/main merge and review follow-ups (400 on malformed client config; normalizeIdpIssuer import consolidation).
* feat(web,core): edit STDIO cwd/env in Server Settings (#1515)
Persist and edit a STDIO server's working directory and environment
variables through the Server Settings modal.
- Add `env` (pair-array, mirrors headers) and `cwd?` to
`InspectorServerSettings`.
- `storedFieldsToInspectorSettings` lifts the stdio config's `env`/`cwd`
into settings so the modal can show them; they stay on `config` for the
transport. Add `envRecordToPairs` / `envPairsToRecord` helpers.
- `env`/`cwd` remain config-owned on disk (the standard mcp.json location):
the `/api/servers` PUT route writes an edited settings.env/cwd back onto
the stored stdio config on a settings-only patch (supporting clear), and a
config-providing PUT keeps config authoritative. `validateSettings`
accepts `env`/`cwd`.
- Server Settings modal: Working Directory field in Options and a dedicated
collapsible Environment Variables section — both STDIO-only.
- Registry server.json import sets no `cwd` (the schema has no
working-directory concept); documented in serverJson.ts.
Tests: serverList converter round-trip, route write-through (set/clear,
config-precedence, non-stdio no-leak, validation), import preservation
(client-config cwd, registry cwd skip), and form/modal component coverage.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(web): preserve config env/cwd when a settings PUT omits them (#1515 review)
Address Claude's review of #1518.
- `/api/servers` PUT now distinguishes an *absent* `env`/`cwd` (preserve the
stored value) from an explicit empty list / blank string (clear). Previously
`validateSettings` coerced a missing `env` to `[]`, so an env-unaware caller
patching only a timeout would silently wipe the server's env/cwd. The
integrated web client always resends the full GET-rehydrated env, so it was
unaffected — this protects the raw HTTP contract.
- `validateSettings` reports `envProvided`/`cwdProvided`; the write-through only
touches `config.env`/`cwd` for fields the caller actually sent.
- On the omit path, rehydrate the untouched env's keychain values onto the
rebuilt entry so the secret reconcile re-persists them rather than sweeping
the blanked-on-disk keys out of the keychain.
- Minor: guard the stdio cast in `mcpConfigToServerEntries` on the transport
type instead of casting unconditionally.
- Test: settings-only PUT omitting env/cwd preserves config keys + cwd + the
keychain secret.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(state): wait for connect refresh before asserting list RPC gate (#1396)
Replace fixed double-microtask flushes in Managed*State connect-path tests with *Change event waits so the capability gate is asserted after refresh completes.
* test(state): fail fast with a readable message if a connect refresh never dispatches its change event (#1396)
The hardened connect-path gate tests now depend on the gated (capability-less /
empty-list) refresh still emitting its *Change event. A future optimization that
skipped that redundant dispatch would make these awaits hang to the vitest
per-test timeout with no clue why.
Extract the per-file waitFor*Change helpers into a shared, generically-typed
waitForChangeEvent() that races the listener against a short timeout and rejects
naming the event that never fired — turning a future hang into a fast, readable
failure. This also removes the duplicated listener/Promise boilerplate across
the five state test files. No production code changes.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(state): cover void-detail event resolution in waitForChangeEvent
Add a case asserting the helper still resolves for a signal-only (void-detail)
event — CustomEvent normalizes an absent detail to null, so the listener fires
and resolves rather than hanging. Closes the coverage gap noted in review.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs(core): correct TypedEventGeneric void-detail comment to null
CustomEvent normalizes an absent detail to null, so listeners receive null
(not undefined) for void events — as the new waitForChangeEvent void case
establishes. Comment-only; no behavior change.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: cliffhall <cliff@futurescale.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web,core): import client configs and registry server.json (#1348)
Wire up the two server-import affordances that were previously todoNoop:
- Import config: source picker (Claude Desktop / Cursor / Cline / VS Code)
read on the backend via a new GET /api/import-source, plus a file-upload
fallback. Per-server review with import/skip on new servers and
overwrite/skip/rename on id conflicts, then a per-server outcome summary.
- Import server.json: parse the MCP registry single-server format
(npm/pypi/oci packages + remotes, env vars, package/runtime args, URL
template vars) into a runnable config, with a file picker and debounced
live validation; Add Server is disabled until validation passes.
Core gains an isomorphic import strategy layer (core/mcp/import) with a
strategy registry, client-config + server.json parsers, a strategy-agnostic
merge, and a dependency-injected source resolver. useServers exposes
importSource(). Modal headers are bold and the redundant server.json section
header is folded into the modal title.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): collapse the server.json textarea into a File Contents disclosure
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): auto-collapse File Contents disclosure after content loads
Once content is pasted or a file is loaded, collapse the File Contents
disclosure after ~1s with Mantine's built-in accordion animation, signaling
the input was accepted and surfacing the validation/env-var sections. The
disclosure becomes controlled by the modal; clearing the textarea re-opens it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): flash File Contents hover highlight before auto-collapse
Before the File Contents disclosure auto-collapses, paint it with its hover
background for 250ms, then run the collapse over a 325ms animation, so the
collapse reads as intentional.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): rename "Import server.json" to "Import from registry config"
Rename the Add Servers menu entry and the modal header.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): rename "Import config" to "Import client config"
Rename the Add Servers menu entry and the modal header.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(web,core): move import orchestration into core helpers + web hooks
Make the import logic reusable by CLI/TUI and keep the modals display-only:
- core/mcp/import: add parseClientConfig (mcpServers→servers fallback),
resolveServerId, selectServerJsonOption, and buildServerConfigForSelection,
so the full import (parse → select → build → merge) is callable from any
environment.
- core/storage: extract the pure validateStoreId into a Node-free store-id
module (store-io re-exports it) so the isomorphic import layer can validate
ids without pulling Node deps into the browser bundle.
- clients/web/hooks: useServerJsonImport and useImportClientConfig own all the
wizard/parse/debounce/apply state; ServerImportConfigModal and
ServerImportJsonModal are now thin display shells over the dumb panel.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): rename "Import client config" to "Import from client config"
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): remove the Cancel button from the registry import modal
The modal header X (and Escape) still dismiss it; drop the redundant Cancel
button (and the panel onCancel prop).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): hide validation results + name override until content is present
In the registry import modal, the Validation Results section and the Server
Name override field stay hidden until something has been pasted/loaded into the
textarea.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): make Validation Results a collapsible disclosure
Wrap the registry import modal's validation results in a Mantine accordion
disclosure (default open), matching the File Contents section.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): Server Name disclosure with read-only derived name + override
Make Server Name a collapsible disclosure containing a read-only "From
configuration" field (the id derived from the server.json name) and the
editable Override field.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): scroll to and highlight a freshly-added server
After a manual add/clone or an import (client config / registry), the new
server's card scrolls into view and draws an animated border (Mantine Border
Animate); clicking anywhere on the card dismisses it.
- App tracks highlightedServerId; addServerHighlighted sets it (edits and
conflict-overwrites don't). Threaded through InspectorView -> ServerListScreen
-> ServerCard.
- ServerCard scrolls into view on highlight and wraps in <BorderAnimate>;
onClick clears the highlight.
- Pin @gfazioli/mantine-border-animate@2.0.3 (Mantine 8 compatible; 2.0.4+
requires Mantine 9).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(web): highlighted card fills its cell and its controls work on first click
- The BorderAnimate wrapper is display:flex, so the card was shrinking to
content width while the border spanned the cell. Stretch the card with
w=100% (and the wrapper).
- Always render BorderAnimate and toggle show/animate instead of conditionally
wrapping: swapping the element type on clear remounted the card, swallowing
the click that triggered it (e.g. the connect toggle fired clear but not
connect, needing a second click). Verified: one click now connects + clears.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(web,core): address review — OCI env, untrack fixture, rename validation
- Untrack the stray inspector-v1-multi.server.json (accidentally committed) and
gitignore root *.server.json fixtures.
- core/mcp/import/serverJson: forward declared env vars into OCI containers via
'-e KEY' docker args (they previously only set the docker CLI process env);
add a test. Clarify buildServerConfig (destructure env out instead of
spread-then-delete) and document value-vs-default env handling.
- useImportClientConfig: validate 'rename' targets inline (invalid charset /
collision with existing, additions, or other renames) and gate the Import
button on canImport; surface the error on the field.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): highlight all servers from a batch import, not just the last
Track highlighted servers as a set of ids instead of one. addServerHighlighted
accumulates each id (a multi-server client-config import highlights them all);
opening an add/import modal starts a fresh batch. Clicking a card clears only
that card's highlight; only the first highlighted card scrolls into view.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(cli): lift config-file headers/timeouts/OAuth into serverSettings (#1482)
The CLI resolved file-sourced servers via resolveServerConfigs(), which
returns bare MCPServerConfig and dropped disk-level headers, timeouts, and
OAuth — CLI connections from a --catalog/--config file silently ignored
them. The TUI already lifted these via mcpConfigToServerEntries().
Extract the TUI's catalog/config -> per-server-settings resolution into a
shared core module (core/mcp/node/servers.ts: loadServerEntries +
selectServerEntry) and route both CLI and TUI through it, so the two share
one file->settings path. --header still overrides the file's headers for
the run while preserving the file's timeouts/OAuth.
- core: add loadServerEntries/selectServerEntry/headersToServerSettings
- tui: tui-servers.ts re-exports the shared core helpers (no logic dup)
- cli: resolve via loadServerEntries + selectServerEntry; drop the
CLI-only conflict check and bare-config path
- tests: core servers.test.ts (settings lifting, header merge/override,
stdio overrides, seed/error/conflict, selection); CLI integration tests
asserting a config header is sent over HTTP and --header overrides it
- docs: CLI README note on file settings being applied
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(core): address #1500 review — dedup overrides, drop dead wrapper, clearer error, spec refresh
- Remove the now-unused resolveLaunchServerConfigs() wrapper (both CLI and
TUI route through loadServerEntries now); migrate its default-catalog
launch tests to compose withDefaultCatalogPath + resolveServerConfigs.
- Deduplicate applyStdioOverrides against config.ts applyOverrides: export
the latter (now trims cwd) and reuse it in servers.ts.
- selectServerEntry's unknown-name error is source-agnostic ("not found"
not "not found in config file") since entries may be a single ad-hoc
default, where "config file" would mislead.
- Refresh specification/ docs (v2_cli_tui_launcher.md,
v2_catalog_launch_config.md): CLI/TUI resolution now via
loadServerEntries/selectServerEntry; mark gap G1 resolved (#1482);
withDefaultConfigPath → withDefaultCatalogPath.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(cli): extend smoke:cli to cover #1482 lift→serverSettings
Adds end-to-end smoke coverage for the config-file headers/timeouts/OAuth
lift introduced in #1482:
- unknown --server now errors source-agnostically ("not found")
- multi-server catalog selection via --server (and the omit-it error)
- --header merge does not break the stdio connect path
- HTTP transport: a config-file `headers` object is lifted onto the wire,
and a CLI --header overrides it (verified against an in-process test
server's recorded requests via a non-blocking spawn)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(cli,tui): normalize --catalog / --config with web; update --help (#1347)
Bring the CLI and TUI in line with the web catalog vs read-only session
model introduced in #1481/#1483, and close the outstanding --help item
from #1347.
Core (core/mcp/node/config.ts):
- Add `catalogPath` to ServerConfigOptions (writable catalog slot).
- `resolveServerSource()` picks the active source: `--catalog` (writable,
seed-if-missing) wins over `--config` (read-only, error-if-missing).
- Writable sources seed an empty `{"mcpServers":{}}` on first run (matches
the web backend) instead of erroring; read-only `--config` still errors.
- `serverSourceConflict()` shared conflict matrix (mirrors run-web):
--catalog/--config mutually exclusive; neither combines with an ad-hoc
target.
- `readServerListFile()` public loader for runners (TUI) needing the raw map.
- Rename `withDefaultConfigPath` → `withDefaultCatalogPath`; the no-flag
default now fills the writable catalog slot, not --config.
CLI / TUI:
- Add `--catalog <path>` (+ MCP_CATALOG_PATH) and re-document `--config` as a
read-only session in --help.
- Enforce the shared conflict matrix before resolution.
Full writable persistence/CRUD threading remains deferred to #1482 / #1432
(CLI/TUI are read consumers today; the split surfaces as seed- vs
error-on-missing).
Tests: core resolution matrix (source/conflict/seed/read-only), CLI catalog
+ conflict integration tests, TUI loadTuiServers unit tests. READMEs updated.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(launcher): add smoke:cli + smoke:tui e2e launcher smokes (#1347)
The launcher smokes only checked `--help` for --cli/--tui; nothing
exercised the launcher → CLI/TUI → core path on the built artifact (only
smoke:web did, for web). Add two end-to-end smokes wired into
validate:launcher between smoke:launcher and smoke:web:
- smoke:cli (scripts/smoke-cli.mjs): runs `--cli` against the bundled
stdio test server via a temp --catalog and asserts tools/list returns
the tools, plus the #1347 resolution paths — default-catalog
seed-on-missing, read-only --config error-without-seed, and the
--catalog/--config conflict.
- smoke:tui (scripts/smoke-tui.mjs): launches `--tui --catalog <temp>`
and asserts the Ink app renders its first frame ("MCP Servers") within
a timeout, then SIGTERMs it (shallow boot/render check).
Both build test-servers/build on demand if missing. Docs updated
(AGENTS.md validate description, launcher README).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor: address review on #1499 (comment accuracy, CLI/TUI flag parity)
Review feedback from @claude on PR #1499 (all non-blocking):
- Align CLI catalogPath resolution with the TUI: `options.catalog?.trim()
|| process.env.MCP_CATALOG_PATH` so an explicit empty `--catalog ""`
falls back to MCP_CATALOG_PATH identically in both front-ends (CLI used
`??`, which only guards undefined/null).
- Clarify the `serverSourceConflict` comment: it mirrors only the
source-selection portion of web's matrix and deliberately omits web's
`--header` rejection, since CLI/TUI merge --header into per-server
settings.
- Note the deliberate --header broadcast semantics in loadTuiServers.
- Share an EMPTY_CATALOG_CONTENT constant between seedEmptyCatalog and
readServerListContent instead of two separate literals.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(cli): trim --config to match --catalog and the TUI (#1499 review)
Cosmetic uniformity follow-up from the re-review: configPath now uses
`options.config?.trim() || undefined`, matching the catalog slot above it
and the TUI. Behavior is unchanged (resolveServerSource/serverSourceConflict
already trim downstream).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): catalog vs read-only session launch model + apply --header (#1481, #1483)
Per specification/v2_catalog_launch_config.md, `mcp-inspector --web` now selects
the backend's server-list source and a session-wide writability, instead of
seeding an inert `initialMcpConfig` the UI never read:
- `--catalog <path>` (and `MCP_CATALOG_PATH`) → active WRITABLE catalog
(seed/CRUD/watch), like the default catalog.
- `--config <path>` → READ-ONLY session file: served in the UI but never
written, seeded, or plaintext-secret-migrated — safe to point at a foreign
config (e.g. claude_desktop_config.json).
- ad-hoc `--server-url`/command (+ `--header`) → one server held IN MEMORY
(no file written); `--header` is lifted onto the entry so the seeded
connection actually uses it (closes the #1483 warn-only no-op).
- nothing → default catalog (writable), unchanged.
Backend (`createRemoteApp`): new `writable` + `initialServers` options. GET
/api/servers branches three ways (in-memory / read-only file with no
seed+no migrate / writable catalog); the four mutating routes return 403 when
read-only; `/api/config` carries `writable`; the file watcher is skipped for
in-memory sessions. `--catalog`/`--config` are mutually exclusive and neither
combines with an ad-hoc target or `--header`; `--header` requires an ad-hoc
HTTP/SSE server (clear errors otherwise — no silent no-op).
UI: `useServerListWritable` reads `writable` from /api/config; the Servers
screen hides add/edit/clone/remove/reorder/settings and shows a read-only
banner when false. All new props default to writable=true (back-compatible).
Docs: spec gaps G2/G3/G4 updated; launcher README documents the flags.
Verified end-to-end against the built launcher: --config (writable:false,
file byte-stable, mutations 403), ad-hoc --header (in-memory, header lifted),
and --catalog (writable:true, POST persists).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(web): address review — Mantine <Code> in read-only banner, clarify watcher comment
- Replace raw <code> with Mantine <Code> in the read-only-session Alert
(repo convention: prefer Mantine primitives over raw HTML).
- Add a comment near `watchable` clarifying that a read-only `--config` file
is still watched on purpose (read-only blocks UI writes, not external
observation; `writable` does not gate watching).
Deferred (per reviewer): consolidating useSandboxUrl + useServerListWritable
into one useConfig hook to avoid the duplicate /api/config fetch — optional
follow-up. The redundant ad-hoc header type-guard is kept as defensive.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(core): make "error" the canonical terminal status on a mid-session crash (#1490)
On a real mid-session transport crash many SDK transports fire BOTH
`onclose` and `onerror` in a transport-dependent order. The `onclose`
guard (`if status !== "disconnected"`) let a trailing `onclose`
downgrade a crash's `"error"` status back to `"disconnected"`, so the
persistent terminal status differed by ordering:
- onerror-then-onclose → "disconnected"
- onclose-then-onerror → "error"
`onclose` now treats `"error"` as terminal (only `"connected"`/
`"connecting"` → `"disconnected"`), so `"error"` is the canonical resting
status in both orderings. It still emits the `disconnect` event exactly
once so session-teardown consumers fire identically either way.
Because the state managers gated list/message/subscription/task teardown
on `getStatus() === "disconnected"`, holding the status at `"error"`
would have left them stale in the onerror-first ordering — swapping one
ordering-dependence for another. Introduce `isTerminalStatus()`
(disconnected || error) and key every teardown guard off it so they
clear identically regardless of crash ordering.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(core): single disconnect event when explicitly disconnecting from "error" (#1490 re-review)
Addresses the non-blocking observation from the PR review: explicitly
calling disconnect() from a held-"error" status could fire the
`disconnect` event twice. close() can trigger the transport's onclose
synchronously; onclose (holding status at "error") fired `disconnect`,
then disconnect()'s own guard saw status still != "disconnected" and
fired it again — a timing-dependence on whether the SDK closes sync.
Add a `disconnecting` ownership flag: onclose defers its status set +
`disconnect` event to disconnect() while it is set, so the
explicit-disconnect path emits exactly one `disconnect` regardless of
sync/async close. Crash-path onclose (flag unset) is unchanged.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(core): deterministically pin the sync-close disconnect double-fire (#1490 re-review)
The real-stdio test asserts the single-disconnect invariant but may not
exercise the sync-close branch (the one that double-fired) if the SDK
stdio transport closes asynchronously. Add a test that overrides the SDK
client's close() to invoke the transport's onclose synchronously,
forcing the exact timing the `disconnecting` flag guards. Verified it
asserts 2 (fails) without the flag and 1 (passes) with it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Closes#1323. Adds an `error`-event-driven `lastError` on useInspectorClient and a mid-session failure toast in App.tsx; the transport onerror is scoped to non-handshake (status !== connecting) so handshake failures stay on the connect() rejection path. Follow-up: #1490.
Ports the CLI, TUI, and launcher from v1.5 into v2 behind a single
`mcp-inspector` entrypoint (launcher → web/cli/tui). Includes the
`mcp-inspector --web --dev` launcher fix (Vite alias resolution + prod-server
double-start), the root `postinstall` install cascade with simplified CI, and
spec/AGENTS docs (no-workspaces rationale, Development setup).
Closes#1246.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(core): trace dropped fetch body update when entry rotates out (#1390)
FetchRequestLogState.onFetchRequestBodyUpdate silently dropped a deferred
response-body update when the matching entry had already been evicted by
log rotation (>= maxFetchRequests newer requests appended between the
append and the fire-and-forget body read). The user saw an empty Response
Body with no way to distinguish "body never came back" from "body dropped
after eviction".
Plumb an optional pino logger into FetchRequestLogState (defaulting to the
silent logger) and emit a logger.debug at the early-return site so the drop
is traceable. The web client wires the existing remote logger from
createWebEnvironment into the state manager.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(core): emit dropped-body trace at warn level so it is observable (#1390)
Review feedback: the web remote logger (createRemoteLogger) defaults to
level 'info' with no override at the environmentFactory call site, so a
logger.debug at the drop site is filtered out by both the pino level gate
and transmit.level — the diagnostic never reaches /api/log and the drop
stays invisible in the shipped web app, defeating the issue's goal.
Promote the trace to logger.warn (>= info, so it transmits by default). A
response body lost to log rotation is unexpected data loss, not routine
diagnostic noise, so warn is the appropriate severity. Update the comment
to explain the level choice and switch the test stub from debug to warn,
extracting a small typed makeWarnLogger helper per review.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): configurable Network Log Size per server + body-dropped toast (#1390)
Turn the otherwise-invisible "response body dropped after rotation" case into
an actionable signal:
1. Per-server `maxFetchRequests` setting (default 1000, 0 = unlimited),
editable as "Network Log Size" in the first (Options) accordion section of
the Server Settings form. Plumbed through the full settings pipeline:
InspectorServerSettings + StoredMCPServer types, the serverList disk
converters (read defaults to 1000, write omits the default to keep diffs
minimal — 0 is a meaningful non-default), and the node server's GET-sanitize
+ write-side validateSettings (non-negative number). Applied live to the
active connection on settings-modal close via a new
FetchRequestLogState.setMaxFetchRequests (shrinking trims immediately), and
at connect time for the next session.
2. FetchRequestLogState now emits a `fetchRequestBodyDropped` event, but only
for genuine rotation drops (log at capacity with a finite cap) — not for
benign post-clear stragglers or unlimited logs, which previously also hit
the warn path. App shows a deduped, persistent toast (stable per-server id,
no auto-close) explaining what happened and the likely causes, with a link
that opens this server's settings on the Options section (modal remounts via
key) so the user can raise the limit if it's just a high-traffic server.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(web): don't treat a cleared Network Log Size field as unlimited (#1390)
Review feedback: clearing the NumberInput emits "", and `parseInt("") || 0`
coerced that to 0 (unlimited). Since the value applies live on modal close, a
user who cleared the field to retype and then closed the modal would silently
switch the log to unlimited. Fall back to the current value on an empty/NaN
parse and reserve 0 for an explicit numeric entry. Adds a cleared-input test.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(web): send a cancellation notification when cancelling an ordinary tool call (#1458)
The Tool detail panel's Cancel button was a silent no-op for ordinary
(non-task) tool calls: App.onCancelToolCall found no captured taskId and
did nothing, leaving the button looking actionable while the call ran on.
Wire request cancellation through the MCP cancellation flow. callTool now
tracks the in-flight call with an AbortController whose signal is threaded
into the SDK request options; aborting it makes the SDK send a
notifications/cancelled for that request and reject the pending call.
cancelToolCall() performs the abort. callTool surfaces the cancellation as
a ToolCallCancelledError (without recording a failed call), which App
clears as a clean cancel plus a short confirmation toast.
Task-augmented calls are unchanged: they have a server-side task and are
still cancelled via cancelRequestorTask (#1457).
Closes#1458
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(web): address review — distinguish disconnect from cancel; tidy callTool params
Review follow-ups on #1463:
- Disconnect no longer surfaces as "Tool call cancelled". The catch in
callToolWithRetries now throws ToolCallCancelledError only when the abort
reason matches the deliberate user-cancel reason (TOOL_CALL_CANCELLED_REASON);
a disconnect aborts the same controller with a different reason and falls
through to the normal error path, so no misleading toast.
- Note the activeToolCallAbortController slot is shared (last-writer-wins)
across all callTool callers.
- Fold callToolWithRetries/attemptToolCall's long positional parameter list into
a single ToolCallRequest object; move the URL-elicitation loop state into
callToolWithRetries.
Adds an integration test that a disconnect mid-call rejects as an ordinary
error (not ToolCallCancelledError), using a slow send_progress call over HTTP.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(web): cancel the underlying task when cancelling a task tool call (#1455)
The ToolDetailPanel Cancel button fired onCancel, but App never supplied
an onCancelToolCall handler, so clicking Cancel did nothing — a tool
running as a task kept running on the server.
App now captures the in-flight task-augmented call's taskId from the
`toolCallTaskUpdated` event callToolStream dispatches, and the new
onCancelToolCall handler cancels that task via the existing tasks API
(reusing onCancelTask). The id is reset at the start of every call, so an
ordinary (non-task) call's Cancel stays a no-op (no task to cancel, no
request-abort channel yet) — preserving prior behavior.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): show a toast when a task is cancelled
Cancellation goes out as a `taskCancelled` event (not a status
notification), so a running task's open 'Task <status>' toast would
otherwise hang with no confirmation. Subscribe to `taskCancelled` and
replace that toast (or show a fresh one) with a short auto-closing
'Task cancelled' toast, pruning the dead progress entry. Covers both
cancel paths — the Tasks screen and the Tool detail panel.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(core): report a cancelled task as cancelled, not failed
When a task is cancelled, the in-flight callToolStream rejects with a
generic -32603 error, whose error path reported the task as "failed" —
so the Tasks screen flashed FAILED until a manual refresh fetched the
server's true "cancelled" state. InspectorClient now records cancelled
task ids (set in cancelRequestorTask before the await, since cancelling
unblocks the stream) and the stream's error path labels the terminal
task "cancelled" with a matching status message. Cleared on disconnect.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor: address review nits on task-cancel wiring
- Prune the consumed taskId from cancelledTaskIds in the stream error path
so the set can't grow across a long session (disconnect-clear remains the
backstop for cancels whose task completed before the cancel landed).
- Clear activeToolCallTaskIdRef before the cancel resolves so a rapid second
Cancel click doesn't re-cancel the terminating task (which would surface a
spurious 'Failed to cancel task' toast). Add a test.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): reorderable ServerCards — drag-and-drop + keyboard (#1369)
Let users control the order of servers on the Servers screen instead of
hand-editing mcp.json. Each ServerCard gets a grip handle (in the header,
left of the name) that drives an @dnd-kit sortable; pointer and keyboard
sensors both work, with live-region announcements for screen readers.
- Backend: PUT /api/servers/order rewrites mcp.json in the supplied order
inside the write lock, reusing the atomic-write + watcher-notify path.
Registered before /:id so "order" isn't captured as a server id. Rejects
with 409 when the id set no longer matches disk and 400 on duplicate /
malformed bodies, so a reorder racing an external edit can't drop or
duplicate an entry.
- Hook: useServers.reorderServers(orderedIds) — optimistic local reorder,
PUT, refresh-to-disk-truth on failure.
- UI: SortableServerCard wrapper owns the dnd-kit concerns; ServerListScreen
wraps the grid in DndContext + SortableContext (rect strategy) with
pointer + keyboard sensors and announcements. Reorder math and a11y copy
extracted to reorderIds.ts / serverReorder.ts for direct unit testing.
Renders plain cards (no grip) when no onReorder is wired.
- Wired onServerReorder through InspectorView -> App.tsx.
Tests: pure-helper units, hook (persist / optimistic / 409 revert), 9
backend integration tests (incl. id-set-mismatch rejection and the
route-registration guard), ServerCard/ServerListScreen units, and a
real-browser Storybook keyboard-reorder play (the dnd-kit keyboard sensor
needs layout rects that happy-dom doesn't provide).
Adds @dnd-kit/core, @dnd-kit/sortable, @dnd-kit/utilities.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(web): notify on reorder failure; skip reorder glue when not reorderable
Address PR #1447 review:
- App.tsx: catch reorderServers rejection (409 from a racing external edit,
or a network error) and show a red toast, matching the error-handling
pattern used by every other mutation — previously the rejection was
discarded with `void` and the drag silently bounced back.
- ServerListScreen: only build handleDragEnd / announcements when
`reorderable`, so the four-closure announcements object isn't allocated
every render when no onReorder is wired.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat: apply auto-refresh setting live + diff-aware list-changed indicator (#1444)
Two follow-ups to #1402.
1. Live setting application — InspectorClient gains setServerSettings()
(on the protocol + fake); App pushes edited settings onto the live
client when the Server Settings modal closes for the connected server,
so toggling auto-refresh-on-list-changed takes effect without a
reconnect (the managed state reads getServerSettings() at notification
time).
2. Diff-aware indicator — in flag-only mode, on */list_changed the managed
Tools/Prompts/Resources state now fetches the new list and compares it
to the current one, lighting the indicator only when it actually
differs. The displayed list stays put until the user pulls via Refresh
(pull-on-demand preserved). Servers like everything re-send identical
lists on list_changed; this suppresses the false-positive indicator.
Refactors each manager's refresh into fetch + apply helpers shared with
the peek path.
Trade-off: item 2 reintroduces a fetch on */list_changed (to compare), but
unlike the pre-#1402 behavior it never replaces the displayed list under
the user — only the indicator reflects a real change.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(core): make the diff-aware peek clear the indicator on revert (#1444 review)
peekForChange now sets the list-changed flag to whether the server list
differs from the displayed one (setListChanged(differs)) instead of only
ever setting it true. So a later notification that reverts the server back
to the displayed list clears the indicator — nothing left to pull. Adds a
clear-on-revert regression test.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(core): consolidate managed-list state managers onto a shared base (#1444)
The four per-primitive list managers (tools, prompts, resources, resource
templates) were near-identical copies of the same fetch/apply/peek/connect/
disconnect/list_changed machinery, a standing drift risk. Extract a generic
ManagedListState<T, M> base parameterized by a config (change event,
list_changed notification, capability key, page fetcher, and whether the
list drives an indicator). Each manager is now a ~40-line subclass: its
config plus a typed getter alias.
Public contracts are unchanged — same per-manager *Change / listChangedChange
events and get*/refresh/getListChanged/clearListChanged methods — so the
hooks and all existing tests pass untouched. Templates opt out of the
indicator (supportsIndicator: false): a list_changed auto-refreshes only
when the server opts in, else does nothing. Net -245 lines.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* perf(core): coalesce bursts of list_changed peeks (#1444 review)
A rapid burst of */list_changed notifications fired one concurrent
paginated fetch per notification. Guard peekForChange with an in-flight
flag: while a peek is fetching, a new notification queues a single re-run
instead of launching a concurrent fetch, and the running peek loops once
more on completion to capture the latest server state. Collapses a burst
into sequential peeks. Adds a coalescing test.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(core): debounce list_changed bursts into a single fetch (#1444)
The in-flight guard only coalesced fetches that actually overlapped — when
a server (e.g. everything) emits a burst of */list_changed spaced just far
enough apart that each list call returns before the next notification, we
still fired one list call per notification. Add a real debounce in the
ManagedListState base: a notification (re)starts a timer and the
refresh/peek runs once the burst settles, collapsing N notifications into
one fetch. The in-flight coalescing guard stays as a second line of defense
for a post-debounce notification landing during a slow paginated fetch.
The delay is injectable (DEFAULT_LIST_CHANGED_DEBOUNCE_MS = 250; subclasses
take an optional constructor arg) so tests run with 0 for fast, determinist
ic behavior. Timer is cleared on disconnect, unsubscribe, and destroy. Adds
debounce + in-flight regression tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(core): extend the in-flight guard to the auto-refresh path (#1444 review)
The overlap guard was only on the peek path; a burst-during-a-slow auto-
refresh could let a slower older fetch clobber a newer list (applyItems is
last-write-wins). Lift the guard from peekForChange up to runListChanged so
it wraps both the auto-refresh and peek actions: while one is fetching, a
new post-debounce notification queues a single re-run. Adds an auto-refresh
coalescing test alongside the peek one.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(core): light the indicator without fetching when auto-refresh is off (#1444)
Per maintainer feedback: with auto-refresh off, a list_changed should not
issue any list call. The diff-aware peek (fetch-to-compare) meant one
tools/list went out per settled burst even with auto-refresh off, which the
maintainer doesn't want. Revert flag-only mode to lighting the indicator
blindly on a (debounced) list_changed — no network — and the user pulls the
new list via Refresh.
Trade-off (accepted): the indicator lights even when the server re-sent an
identical list (the everything-server case), since we no longer fetch to
compare. Keeps the debounce (one indicator light per burst), the
auto-refresh path + its in-flight guard, the live setServerSettings, and the
ManagedListState consolidation. Removes peekForChange and its tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs(core): scrub stale peek docstrings after the blind-indicator revert (#1445 review)
The diff-aware revert removed peekForChange's body but left its docstring,
which orphaned onto destroy() and described fetch-to-compare/clear-on-revert
behavior that no longer exists. Replace it with a one-line destroy() doc, and
update clearListChanged's doc (peek -> blind light).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs(core): finish scrubbing stale peek references after the revert (#1445 review)
Round-7's scrub missed four more peek/fetch-to-compare mentions in
managedListState.ts (supportsIndicator, fetchAll, the debounce-default and
onListChanged comments) plus six 'async peek' comments in the state tests.
Update them all to the blind-indicator semantics. Doc/comment-only; no
behavior change. grep -i peek over core/mcp/state now returns nothing.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): wire the listChanged indicator across all screens (#1402)
The per-screen list-changed indicator was hardcoded false in
InspectorView for Tools, Apps, Prompts, and Resources, so it could
never light up. Source it from the managed-state layer instead.
Each ManagedXState now tracks a listChanged flag: set true when its
*/list_changed notification arrives (the existing auto-refresh leaves
it set), exposed via getListChanged()/a listChangedChange event, and
cleared by clearListChanged() or on disconnect. The useManagedX hooks
surface the flag and clear it on a user-initiated refresh (the
indicator's Refresh button = acknowledge). App threads the three flags
to InspectorView, which maps them to the screens — Apps shares the
tools flag since apps are a filtered view of tools.
Tests cover set/clear/disconnect at the state-manager layer, the hook
subscription + refresh-clears behavior, and the InspectorView routing
(including Apps sharing the tools flag and no cross-wiring between
screens).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(core): close clear-after-await race in managed-list refresh (#1402 review)
The useManagedX refresh() callbacks cleared the listChanged flag AFTER
awaiting the fetch. If a */list_changed arrived mid-fetch, the state
re-set the flag (and auto-refreshed), then the trailing clear wiped that
genuinely-new signal. Clear BEFORE the await instead: the user's refresh
acknowledges only the change in hand, and a mid-fetch change re-sets the
flag and survives. Adds a race regression test on the tools hook.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(web): drop duplicate Refresh on the Apps screen (#1402)
AppControls uniquely had a standalone toolbar Refresh button AND the
ListChangedIndicator on a separate row, so once the indicator could
actually appear (this PR), the Apps screen showed two Refresh buttons.
Match the other three controls (Tools/Prompts/Resources): the indicator
is the single header-right refresh affordance. Updates AppControls and
AppsScreen tests to assert exactly one Refresh (and none when the list
hasn't changed).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(core): stop auto-invoking */list on list_changed; user pulls via Refresh (#1402)
With auto-refresh on, the list was already current by the time the
list-changed indicator appeared, making its Refresh button moot. Now the
managed Tools/Prompts/Resources state managers only flag the change on
*/list_changed (they no longer auto-refresh); the list stays as-is until
the user clicks Refresh, which fetches the new list and clears the
indicator. The connect-time initial load is unchanged.
Resource templates (no own indicator, part of the Resources screen) also
stop auto-refreshing; onRefreshResources now refreshes both resources and
templates, since a single resources/list_changed covers both. Tasks are
left auto-refreshing — they have no list-changed indicator and need live
status updates.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs(core): correct ManagedResourceTemplatesState header comments (#1402 review)
The class no longer subscribes to resourceTemplatesListChanged; update
the stale file/class docstrings that still claimed it syncs on that
notification to reflect connect + statusChange only.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): add per-server 'Auto Refresh on List Changed Notifications' option (#1402)
New Options section in ServerSettingsForm with a single checkbox. When
checked, the managed Tools/Prompts/Resources (and resource templates)
state auto-refreshes on */list_changed notifications; when unchecked
(default) the notification only lights the list-changed indicator and the
user pulls via Refresh.
Plumbing: new optional autoRefreshOnListChanged on InspectorServerSettings
and StoredMCPServer, round-tripped through serverList (lifted with a false
default, persisted only when true, added to the exhaustive field-key map);
surfaced from InspectorClient via a new getServerSettings() on the protocol;
the state managers read it live at notification time to decide auto-refresh
vs flag-only. Resource templates re-subscribe to their list_changed but only
auto-refresh when the option is on (they have no indicator of their own).
Tests cover the round-trip (true persists, false omitted), the form
section (render/checked/toggle), and each state manager's on/off branch.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(server): persist autoRefreshOnListChanged through validateSettings (#1402)
The PUT /api/servers write path runs settings through validateSettings,
which rebuilds the value from explicitly-named fields and silently drops
unknown keys — so autoRefreshOnListChanged never reached disk (the
serverList round-trip test passed because it calls
inspectorSettingsToStoredFields directly, bypassing this path). Validate
it as an optional boolean and include it in the built value; false is
omitted on disk by inspectorSettingsToStoredFields as before.
Adds integration tests over the real PUT path: true persists, false is
omitted, and a non-boolean is rejected (400).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): default the Server Settings modal to the Options section (#1402)
Open 'Options' instead of 'Custom Headers' when the Server Settings modal
first opens, so the new Auto Refresh toggle is the first thing surfaced.
Updates the modal default-expanded section, the modal test's initial-state
assertions, the three header tests (now expand 'Custom Headers' first since
it's no longer open by default), and the story defaults.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* chore: drop accidentally-committed Inspector history dump; gitignore the class
83076441 swept an exported Inspector message-history JSON (manual-testing
artifact at the repo root) into the tree via git add -A. Untrack it and
extend the existing repo-root debug-artifact .gitignore block (alongside
/*.png) to cover /inspector-history-*.json so these can't be re-added.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): show negotiated MCP protocol version on ServerCard (#1324)
Add optional protocolVersion to ConnectionState and render it as a
dimmed "MCP <version>" label at the end of the transport-badge/
description row on ServerCard, shown only when connected. The value
is plumbed through useInspectorClient as part of #1324.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(core,web): surface negotiated protocolVersion end-to-end (#1324)
Plumb the MCP protocol version negotiated during initialize from the
InspectorClient through to the ServerCard so the active connected card
displays "MCP <version>".
The SDK Client doesn't expose protocolVersion; it stamps it onto the
transport via setProtocolVersion. MessageTrackingTransport now captures
it (and forwards to the base transport when present), so it works for
stdio too — whose base transport has no setProtocolVersion.
- core: MessageTrackingTransport captures + exposes protocolVersion;
InspectorClient stores it, exposes getProtocolVersion(), dispatches a
protocolVersionChange event, and clears it on disconnect.
- core: useInspectorClient subscribes and returns protocolVersion.
- web: App reads it from the hook (drops the hard-coded "2025-06-18")
and InspectorView splices it onto the active server's connection.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(web): decouple initializeResult from protocolVersion (#1324 review)
Gate the initializeResult memo on serverInfo only, not protocolVersion.
initializeResult also drives the connected ViewHeader and Connection Info
modal, so requiring the version risked hiding the whole connected surface
in the edge case where it's absent. protocolVersion now flows through as
the optional field it already is everywhere downstream ('' reads as
unknown; the ServerCard label and modal value both tolerate it).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(web): drop redundant connected-check in InspectorView splice (#1324 review)
initializeResult is derived from connectionStatus in App's memo, so a
truthy initializeResult.protocolVersion already implies a live
connection — the extra connectionStatus === 'connected' guard in the
splice was redundant. ServerCard keeps its own status guard (dumb-
component robustness). No behavior change.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(web): don't show a permanent PENDING badge on history notifications (#1439)
`extractStatus` derived the history status badge purely from whether a
`response` was attached, so a fire-and-forget notification (no id, no
response ever) — and any unmatched standalone response — was labelled
"Pending" forever. The pending → OK/Error lifecycle only applies to
request entries (messageLogState attaches the response by JSON-RPC id),
so scope the status to `direction === "request"` and render no
request-style badge for notifications/responses. The method badge already
labels them.
Part of the History-screen work on #1438.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(core): track outgoing responses so server→client requests resolve in history (#1440)
MessageTrackingTransport.send only tracked outgoing requests, so the
client's response to a server→client request (roots/list, sampling,
elicitation) was never recorded. The inbound request entry never got its
response folded in, leaving it stuck "Pending" with no response body.
Make send symmetric with onmessage: track outgoing responses (by id +
result/error) as well as outgoing requests. The response now correlates
to its request entry, which resolves to OK/Error with the body. Outgoing
notifications remain untracked (separate gap).
Part of the History-screen work on #1438.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): wire up History Pin and Replay (#1438)
Both HistoryEntry actions were rendered but stubbed to todoNoop in App.
- Pin: App now owns a session-scoped Set<string> of pinned entry ids,
toggled by onTogglePinHistory and passed down as pinnedHistoryIds.
HistoryListPanel already sorts pinned entries to the top. The set resets
with the rest of the per-screen state on disconnect/server-switch.
- Replay: onReplayHistory re-issues the entry's original request by method
(tools/call → callTool, prompts/get → getPrompt, resources/read →
readResource). The call flows through InspectorClient → tracked
transport → message log, so the fresh request+response surface as a new
History entry (history-local) without touching the Tools/Prompts/
Resources panels. Unsupported methods / a removed tool surface a toast.
Tests: pin toggle propagation, replay dispatch for each supported method,
unsupported-method and missing-tool toasts.
Closes#1438.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): make history Replay support list/discovery requests
Replaying a tools/list entry reported "Replay isn't supported". Extend
replayHistoryRequest beyond the call-style requests to the read/discovery
methods, re-issuing each via its InspectorClient method (preserving the
pagination cursor):
- tools/list → listTools, prompts/list → listPrompts,
resources/list → listResources,
resources/templates/list → listResourceTemplates, ping → ping.
Tests: tools/list replay (cursor preserved) and an unsupported-method
toast (logging/setLevel).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): replay tasks/list; reorder + conditionally show History buttons
- Add tasks/list to the replayable set (re-issued via listRequestorTasks),
completing the client-issued */list methods.
- Reverse the HistoryEntry actions to Pin then Replay, and hide Replay for
entries whose method can't be replayed.
- Introduce a single source of truth — REPLAYABLE_HISTORY_METHODS /
isReplayableHistoryMethod in historyUtils — used by HistoryEntry to
show/hide the button and by App's replayHistoryRequest to gate dispatch,
so the two can't drift.
Tests: tasks/list replay, Replay hidden for a non-replayable method, and
Pin-before-Replay ordering.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): add client↔server direction badge to History and Network (#1437)
A dual-state badge at the top-left of each entry shows which way it
traveled: "client → server" (green) for inspector-originated, "client ←
server" (yellow) for server-originated.
- New shared element MessageDirectionBadge (Elements).
- History: MessageEntry gains an `origin` ("client" | "server") set at
tracking time — MessageTrackingTransport tags `send` as client and
`onmessage` as server, threaded through the track callbacks. HistoryEntry
maps origin → badge. (direction stays the request/response/notification
message-type used for response correlation; origin is the flow.)
- Network: fetches are always inspector-originated, so NetworkEntry renders
the badge as outgoing.
Tests + stories for the badge; HistoryEntry direction tests; transport
test asserts the origin argument.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* style(web): use purple (not yellow) for the client ← server direction badge
Yellow + green read as caution/ok status; purple vs green reads as
direction. Incoming (server → client) is now violet; outgoing stays green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): make History Pinned/History sections collapsible toggles
The "Pinned Requests (N)" and "History (N)" section headers are now
clickable toggle buttons that expand/collapse their entries — styled like
the LogControls level toggles (UnstyledButton listItem variant with the
active background) and wrapping the entries in a Collapse. Both default
open and collapse independently.
Tests assert the per-section aria-expanded toggling and independence.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): replace Expand/Collapse text button with an icon toggle
History, Network, and Task cards now use a shared ExpandToggle ActionIcon
instead of the "Expand"/"Collapse" text button: LuArrowDownToLine when
collapsed (expand), LuArrowUpFromLine when expanded (collapse). The
aria-label stays "Expand"/"Collapse" so the control reads the same.
- New shared element ExpandToggle (+ test + stories).
- HistoryEntry: actions split into a left group (Pin/Replay) with the
toggle pushed right via space-between.
- NetworkEntry / TaskCard: drop their now-unused SubtleButton const.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): pin icon toggle on HistoryEntry, moved to the right
Replace the Pin/Unpin text button with a shared PinToggle ActionIcon
(TiPinOutline when unpinned, TiPin when pinned; aria-label stays
Pin/Unpin). Move it to the right action group, just left of the
expand/collapse toggle — Replay stays on the left.
- New shared element PinToggle (+ test + stories).
- HistoryEntry action row: Replay left; Pin + expand toggle right.
- Drop the now-unused formatPinLabel helper.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): per-section Clear/Export on the Pinned and History headers
Each section header now shows Clear and Export links on the right, but
only when both sections are present (a single section is already covered
by the panel-level Clear/Export). They act on just that section's entries:
- Clear: removes that section's entries by pin membership (clearing the
Pinned section also drops the now-stale pinned id set).
- Export: downloads just that section's entries
(inspector-history-pinned/-unpinned-...json).
Threaded onClearSection/onExportSection App → InspectorView → HistoryScreen
→ HistoryListPanel; App computes each section from messages + pinned ids.
Extracted a CollapsibleSection subcomponent and a SectionActions links
group. Extended ExportKind with history-pinned/history-unpinned.
Tests cover both-sections visibility gating and per-section dispatch.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): Filter by Message Direction on the History controls
Add a "Filter by Message Direction" section to HistoryControls, modeled on
LogControls' "Filter by Level": a Select/Deselect All control plus two
listItem toggles — "client → server" and "client ← server" — that show or
hide entries by origin.
- HistoryUiState gains visibleDirections (Record<MessageOrigin, boolean>),
defaulting to both visible; persists with the other per-screen state.
- HistoryScreen wires the toggles (single + all) through onUiChange and
passes the filter to HistoryListPanel, where matchesFilters drops entries
whose direction is off (entries without a recorded origin are never
filtered out).
Tests cover the controls (render, single toggle, select/deselect all), the
HistoryScreen onUiChange wiring, and the list-level direction filtering.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* style(web): section Clear/Export as subtle buttons, not link anchors
Match the Select/Deselect All control in HistoryControls: the per-section
Clear/Export on the History/Pinned headers now use a subtle Button
(variant subtle, size xs) instead of an Anchor link.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* style(web): color the direction filter labels to match the badge
The "client → server" / "client ← server" toggles in HistoryControls now
color their labels green / violet, matching the MessageDirectionBadge.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): track outgoing client notifications in history too
Review follow-up (#1441): MessageTrackingTransport.send now also tracks
the client's own notifications (initialized, progress,
roots/list_changed) as origin "client", making send fully symmetric with
onmessage. Previously only id-bearing outgoing messages (requests,
responses) were tracked, so client notifications never reached History
while incoming server notifications did.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): Filter by Message Direction on the Network screen
Add the same "Filter by Message Direction" section to NetworkControls as
on HistoryControls. Extracted the shared section into a MessageDirectionFilter
group component (used by both HistoryControls and NetworkControls) so they
stay identical — a Select/Deselect All control plus green/violet
client → server / client ← server toggles.
- NetworkUiState gains visibleDirections (default both visible), persisted
with the rest of the per-screen state; NetworkScreen wires the toggles and
NetworkStreamPanel filters on them.
- Network fetches are always inspector-originated, so all entries are
client → server: the client toggle gates them all and the server toggle
matches nothing (documented in NetworkStreamPanel). The section mirrors
History for parity.
Tests cover the shared filter, both controls' direction sections, the
Network list filtering, and the NetworkScreen wiring; ambiguous "Deselect
All" selectors (now two sections) disambiguated by order.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* revert(web): drop message direction from the Network tab
Network traffic is always client → server — even SSE is a client-initiated
stream tagged outgoing — so a direction badge/filter there is meaningless.
Remove the MessageDirectionBadge from NetworkEntry and the Filter by Message
Direction section (and its NetworkUiState.visibleDirections wiring) from the
Network screen.
History keeps its direction badge + filter; the shared MessageDirectionBadge
and MessageDirectionFilter components stay (still used by History).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(web): a lone History section isn't a collapsible accordion
When only one of Pinned / History is present, its header is now a plain
title and its entries always show — instead of a `listItem` accordion
toggle that could hide them based on a stale collapsed state from when
both sections were on screen. The accordion (and per-section Clear/Export)
only applies when both sections are present.
Example fixed: collapse History while pins exist, then unpin everything —
History becomes the sole section and shows its entries under a plain
header rather than staying collapsed.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* style(web): put section Clear/Export on the header pleat itself
The per-section Clear/Export sat to the right of the collapsible header
bar, on the panel background. Make the whole header row the "pleat":
a new Group `sectionHeader` theme variant (rounded + hover, active bg
passed per instance) wraps a flex-1 toggle button (the title) on the left
and the Clear/Export buttons on the right — so the actions live on the
pleat. The toggle is a separate button since the action buttons can't nest
inside it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* style(web): entry expand toggle uses the same icons as ListToggle
Swap ExpandToggle (History / Network / Task cards) from the LuArrow icons
to RiExpandVerticalLine / RiCollapseVerticalLine — the same expand/collapse
icons the list-level ListToggle uses. Same ActionIcon size, so it matches
the other small buttons on the cards; aria-labels unchanged.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(web): panel Clear keeps pinned entries (history-only)
Review follow-up (#1441): the panel Clear button is gated on the unpinned
count (implying history-only), but onClearHistory wiped everything
(pinned included) and left pinnedHistoryIds stale. Make Clear history-only
— clearMessages((m) => !pinnedHistoryIds.has(m.id)) — so pinning protects
an entry, matching the gating and the per-section model; the pin set stays
valid since its entries survive.
Also drop the stale "and Network" mention from the MessageDirectionFilter
docstring (Network direction filter was reverted).
Test: panel Clear's predicate keeps pinned ids and removes unpinned.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): two-step URL elicitation completion (#1415)
Replace accept-on-open with an explicit completion step for URL
elicitations. Opening the URL now only transitions the panel to its
waiting state; the elicitation resolves with `accept` only when the user
clicks the revealed "I've completed it" action, or with `cancel`.
- ElicitationUrlPanel: add `onComplete`, reveal the completion button and
relabel open→"Reopen in Browser" while `isWaiting`.
- PendingClientRequestModal: wire local `isWaiting`; open no longer sends
a response. Removed the accept-on-open comment that pointed here.
- Tests cover open→waiting→complete and open→cancel, and assert opening
alone does not resolve the elicitation.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): handle URL-elicitation-required (-32042) error path
Implements the spec's "URL mode with elicitation required error" flow
(2025-11-25). When a tools/call returns a `-32042`
UrlElicitationRequired error, the client now surfaces the required URL
elicitation(s) and retries the call once completed, instead of showing a
bare error.
Core (InspectorClient.callTool):
- Detect `-32042` (typed UrlElicitationRequiredError or generic McpError)
via the new `getUrlElicitationsFromError` helper.
- Surface each carried elicitation in order through the existing pending-
elicitation queue / modal (reusing the two-step "Open → I've completed
it" UI), then re-issue the original call once all are accepted. A
decline/cancel aborts with a clear message; a bounded retry count guards
a server that keeps returning the error.
- `ElicitationCreateMessage.completeIfPending()` lets the optional
`notifications/elicitation/complete` auto-advance/auto-accept an open
URL elicitation; the completion-notification handler now resolves
(was: only removed) the pending entry.
- Extracted `attemptToolCall` / `dispatchFailedToolCall` from the old
callTool body to support the retry loop without duplicating bookkeeping.
Web (App):
- A non-spec `-32042` with no elicitations (nothing to open) shows a
dedicated yellow toast linking to a new UrlElicitationErrorModal with
the raw error body, mirroring the output-schema-validation toast.
Tests: helper unit tests, completeIfPending unit tests, callTool
error-path retry/cancel/no-list/ordinary-error tests (injected fake
client), the new modal (test + stories), and an App toast test.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): abort URL-elicitation error path on a repeated URL
Guard the -32042 retry loop: if the server's retry error re-requests a
URL the user already completed earlier in the same call, completing it
again can't make progress, so callTool aborts with a new typed
UrlElicitationLoopError instead of re-prompting (which would loop until
the retry cap). The web layer detects it and shows a "URL elicitation
loop" toast naming the repeated URL rather than a generic failure.
Tests: core abort-on-repeat test (URL presented once, two attempts, then
loop error) and an App loop-toast test.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(web): let Copy-URL reach the URL-elicitation completion step
Address review of #1436: the "I've completed it" action was gated behind
clicking "Open in Browser" (only that set isWaiting), so a user who used
Copy URL to finish the flow externally could only Cancel. Copying the URL
now also moves the panel into the waiting state, exposing the completion
step. Also trims the now-redundant second sentence of the waiting hint.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(core): address review of error-path URL elicitation
Review follow-ups on #1436 (all non-blocking):
- Remove the leftover bare block scope in attemptToolCall (extraction
artifact); body de-indented one level.
- Log a warning when callTool gives up after the URL-elicitation retry
cap is hit, so a server that keeps demanding new elicitations is
diagnosable rather than looking like an ordinary failure.
- Assert the dispatch-once invariant on the loop-abort and cancel-abort
paths (exactly one failed toolCallResultChange event).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* style(core): fix attemptToolCall body indentation
Run prettier over the extracted attemptToolCall method body, which was left
over-indented by 2 spaces after the bare-block removal in 0754b43d. core/ is
outside clients/web's format:check scope, so npm run validate did not catch it.
Whitespace-only; the pre-existing line 658 wrap (also outside the format scope)
is left as-is, out of scope for this PR.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(core): settle pending elicitations on disconnect to avoid callTool hang
Address review of #1436: an error-path `awaitUrlElicitation` promise only
settled via respond, but `disconnect()` cleared `pendingElicitations`
wholesale — so a disconnect mid-elicitation (server crash, manual
disconnect, transport close) left `callTool` (and the UI's in-flight
toolCallState) hanging forever.
- Add `ElicitationCreateMessage.cancel()`: settles a pending elicitation
as `{ action: "cancel" }` without calling `onRemove` (the queue is
cleared by the caller, so it must not splice mid-iteration).
- `disconnect()` now cancels each pending elicitation before clearing the
array, and dispatches `pendingElicitationsChange` (it already did so for
samples) so the modal closes on a transport-close teardown too.
Tests: disconnect-mid-elicitation settles callTool (no hang); cancel()
unit tests (resolves-without-remove, no-op once responded).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): add per-server Roots config + client roots capability (#1425)
Add MCP roots support to the v2 web Inspector as a per-server config
section in the server settings form (no separate Roots tab/screen).
- Persistence: add `roots: Root[]` to `InspectorServerSettings` and
`StoredMCPServer`; thread through serverList converters (lift on read,
drop empty-uri rows / blank names on write, omit when empty) and the
remote server's read-side drop guard + `validateSettings`.
- Capability: `setupClientForServer` always advertises the roots
capability (`roots: cleanRoots(savedSettings?.roots ?? [])`) so the
server can issue `roots/list` and receive `roots/list_changed`.
- Live apply: edits to a connected server's roots are pushed via
`inspectorClient.setRoots(...)` once, on settings-dialog close — diffed
against the client's current roots — rather than per keystroke, to
avoid flooding the server with `roots/list_changed` / `roots/list`.
- Form: new Roots accordion section with a required URI and an optional
name per row (unlike v1, which was URI-only), + Add Root, and remove.
Tests cover the persistence round-trip (serverList + servers-route),
the form/modal section, and reuse the existing InspectorClient
roots/`setRoots` integration coverage.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(roots): address review — dedup guards/cleanRoots, preserve extra fields, test live-apply
Responding to the PR #1431 code review:
- Dedup `isRootArray`: drop the copy inside `validateSettings` and reuse the
module-scope guard the read path uses, so the two can't drift
(core/mcp/remote/node/server.ts).
- Dedup the clean-roots normalizer: hoist a single exported `cleanRoots`
into core/mcp/serverList.ts and reuse it from both
`inspectorSettingsToStoredFields` and App.tsx (drops the local copy).
- Preserve non-form root fields (e.g. `_meta`): `cleanRoots` now spreads the
carried-through fields and only normalizes uri/name, and the modal's
`handleRootChange` spreads the existing root on edit — so a hand-edited
root's `_meta` survives load + persist + edit instead of being silently
dropped.
- Test the App live-apply-on-close path: new App.test.tsx cases assert
setRoots is called once with cleaned roots when changed on the active
server, and not called when the cleaned roots already match what the
client advertises. Adds a cleanRoots unit test (blank drop + _meta
preserve).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Wires the ported Tasks screen + core task subsystem through App.tsx: sticky Clear Completed, cancel/refresh error toasts, core-side per-task progress correlation, Run-as-task on the Tools screen (gated on server support), per-server Task TTL, and live task-status toasts.
* feat(web): render pending sampling/elicitation requests (#1407)
Server-initiated sampling and elicitation requests never reached the v2
web UI: the InspectorClient queued them and returned a Promise that only
resolved on a user response, but nothing subscribed to the pending-change
events or rendered the (already-built) request components. A tool that
triggered one spun the Execute button forever.
Wire the queues to the UI:
- Add usePendingClientRequests() (core/react) subscribing to
pendingSamplesChange / pendingElicitationsChange and returning the live
arrays. Expose getPendingSamples() / getPendingElicitations() on
InspectorClientProtocol (and the test fake) so hooks can read them.
- Add PendingClientRequestModal, an App-level modal that surfaces one
queued request at a time (sampling-first) via the existing
SamplingRequestPanel / ElicitationFormPanel / ElicitationUrlPanel. It is
non-dismissable so the blocked call only resolves via an explicit action.
- Wire it in App.tsx: responding/rejecting resolves the client's handler
Promise, which unblocks the originating call and clears the spinner.
Also fold in two issues surfaced while verifying:
- SchemaForm rendered an `array` + `items.enum` (multi-select enum) as a
JSON textarea because only `items.anyOf` had a branch; add a MultiSelect
branch with enumNames label support.
- ToolsScreen's result lives in App and outlived the screen's local
selection (reset on tab-switch unmount), leaving a result with no
selected tool. Clear the result on unmount for a clean slate on return.
Prompts/Resources gate their preview on selection, so they are unaffected.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(elicitation): support explicit decline; honest queue label; review polish
Address @claude review on #1411:
- Add an explicit "Decline" action to ElicitationFormPanel so the spec's
decline (refusal) is distinguishable from cancel (dismissal); the JSDoc
claimed accept/decline/cancel but only accept/cancel were wired. Wire the
modal body to emit { action: "decline" }.
- Replace the misleading "1 of N" queue label (the leading 1 never changed
since the modal always shows the head) with an "N pending" hint, hidden when
nothing else is queued.
- Make the dead useMemo honest: derive the head request inside the memo from
the source arrays so it actually caches (the inline activeElicitation had a
fresh identity every render).
- Document the URL elicitation accept-on-open semantic with a comment pointing
at the follow-up (#1415).
- Add an App-level round-trip test driving pendingSamplesChange end-to-end
(modal opens, respond() fires, empty event closes it).
Follow-ups filed: #1415 (URL two-step completion flow), #1416 (form required-
field submit gate + double-submit guard).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(elicitation): include untouched schema defaults in submitted content
The schema form displays a field's default via resolveValue(), but onChange
only writes a field into `values` once the user edits it. So an elicitation
submitted with default-only fields left untouched dropped them entirely —
v2 sent just the manually-edited fields while v1 sent defaults too (reported
with the everything server: firstLine/integer/number/enum defaults missing).
Add collectSchemaDefaults() to jsonUtils (recurses nested objects, omits
fields with no default) and seed ElicitationFormModalBody's values state with
it via a lazy initializer. The modal body is keyed by request id, so it
re-seeds per request; user edits still override seeded defaults via the
existing onChange merge.
Tests: collectSchemaDefaults unit tests (field types, nested, empty) in
jsonUtils.test; a modal test asserting an untouched-default form still submits
its defaults in content.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(forms): seed schema defaults for tool-call and app-launch forms too
Extend the elicitation defaults fix (collectSchemaDefaults) to the other two
SchemaForm consumers that own form state and submit it: ToolsScreen (tool
execute args) and AppsScreen (app launch args). Both reset formValues to {} on
selection, so default-only fields the user never edited were dropped from the
submitted args — same root cause as the elicitation modal.
Seed formValues from the selected tool/app inputSchema defaults on select.
User edits still override via the existing onFormChange merge.
(InlineElicitationRequest has no consumers wired up, so it's left as-is.)
Tests: ToolsScreen sends a tool's default on Execute without editing;
AppsScreen sends an app's default on Open App without editing.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(elicitation): gate form submit on required fields; guard double-submit (#1416)
Fold the #1416 hardening into this PR:
- Gate the elicitation form's Submit button on required fields. Promote the
hasMissingRequiredFields helper out of AppDetailPanel into jsonUtils (shared)
and use it in ElicitationFormPanel so Submit is disabled until every required
field has a value.
- Guard against double-submit. Add an optional `busy` prop to the sampling,
elicitation-form, and elicitation-url panels that disables their responding
actions. PendingClientRequestModal's bodies use a new ref-guarded
useRespondOnce() hook: the first action dispatches and locks the panel; later
clicks no-op. This stops a fast double-click from calling respond()/reject()
twice (which throws and produced an unhandled rejection). Copy URL stays
enabled — it doesn't resolve the request.
Tests: hasMissingRequiredFields unit tests; ElicitationFormPanel submit-gate
and busy-lock tests; a modal double-click-resolves-once test.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(tools): cap detail/result panels at viewport and scroll internally
The Tools screen's detail and result columns sized to their content and bled
past the bottom of the viewport on long output (e.g. the env dump or a long
elicitation result). Cap each column at the screen's available height and let
its panel scroll internally instead.
- ToolsScreen: drop `align: flex-start` and wrap each column in a stretched
`ContentPane` (Flex column, capped by `mah` = viewport − header − xl*2). The
pane's definite height is what lets the card's inner ScrollArea know how far
to shrink — a bare `mah` on the card does not (the card sizes to content, so
the scroll region renders full-height and overflow:hidden just clips it,
unreachable). Mirrors the proven Prompts/Resources preview panes.
- ToolResultPanel: pinned header + inner ScrollArea (flex 0 1 auto, mih 0).
- ToolDetailPanel: pinned title/annotations header, scrolling body
(description + form + progress), pinned Execute footer.
Audited the other screens: Prompts/Resources already use this pattern,
ServerList and the four list panels (Logging/Network/History/Tasks) already cap
via ScrollArea.Autosize. Only AppsScreen's app-input form had the same gap —
fixed it the same way: AppDetailPanel now scrolls internally and the card uses
variant="preview" so the running app's iframe and a long form both stay within
the viewport.
Verified in-browser against the everything server: the get-env result now
stops at the viewport bottom with margin and scrolls within the card
(clientHeight 502 vs scrollHeight 1931), header pinned.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(remote): use sync pino destinations in logging tests to fix teardown race
CI flaked with an uncaught `ENOENT: ... /tmp/inspector-log-unknown-level-*/app.log`
that failed an unrelated later test. The remote-logging integration tests create
`pino.destination(logFile)` loggers (async sonic-boom streams) and rmSync the temp
dir in finally/afterEach without closing the stream — a buffered background flush
then fires after the dir is gone and throws an uncaught error, which Vitest counts
as a run failure even though all tests pass.
Make those destinations `sync: true` so the write lands during the request instead
of on a later flush, removing the race. Test-only; unrelated to this PR's feature
work but it was blocking the PR's CI.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(storage): await store-file writes instead of polling (#1412)
OAuth persistence is Zustand `persist` → file-storage adapter →
writeStoreFile, which the persist middleware invokes fire-and-forget:
saveTokens updates the in-memory store synchronously while the file
write lags. Tests verified persistence by polling the state file with a
2s/50ms `waitForStateFile()` budget, which flaked under full-suite load
(the file was written, the poll just didn't observe it in time).
Make the write completion awaitable instead:
- Track in-flight writeStoreFile() promises per resolved path in
store-io and add flushStoreFileWrites(filePath?) to await them (also
useful for graceful shutdown). Writes to the same path are chained so a
flush awaits every queued write.
- Convert the OAuth E2E, node storage, and file-storage-adapter tests
from poll loops (waitForStateFile / vi.waitFor on existsSync /
setTimeout) to `await flushStoreFileWrites(path)` + a single read.
- Remove the now-unused waitForStateFile() test helper.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* chore(storage): doc load-bearing sync registration; rename stale debug env
Address @claude review on #1413:
- Document that writeStoreFile()'s synchronous pendingWrites registration is
load-bearing for flushStoreFileWrites(), in both the storage adapter's
setItem callsite and the pendingWrites docblock, so future refactors don't
introduce an await before registration and silently re-flake the tests.
- Rename the stale DEBUG_WAIT_FOR_STATE_FILE test env var to DEBUG_PERSIST_PATH
(waitForStateFile was removed in this PR).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(apps): render running MCP Apps end-to-end (#1370)
Wire the v2 Apps screen so clicking Open App actually runs an MCP App.
Previously App.tsx stubbed the three load-bearing pieces (sandbox URL,
bridge factory, action handlers), so the iframe loaded about:blank and
nothing crossed the bridge.
- Add useSandboxUrl hook (core/react) to read sandboxUrl from GET
/api/config and thread it down as sandboxPath; render an "MCP Apps
unavailable" state instead of a blank iframe when absent.
- Add createAppBridgeFactory: builds a host-side ext-apps AppBridge over
the active client's SDK client, reads the tool's UI resource on the
sandbox handshake and pushes it into the proxy, handles openLinks, and
connects a PostMessageTransport.
- AppRenderer: pass the tool to the bridge factory; buffer tool
input/result and release them once the view signals `initialized`
(input before result, latest-wins).
- Move the AppRenderer handle ref to App.tsx so onOpenApp issues
tools/call with the form values and delivers input + result to the
running app via the bridge; onCloseApp tears the bridge/iframe down.
- Stories: themed sandbox placeholder (visible in light/dark), an
echo-bridge running story, and an unavailable story. New unit tests
for the factory, the buffering, useSandboxUrl, and the empty state.
The listChanged indicator badge wiring is tracked separately in #1402.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(apps): make MCP Apps render in dev — single bridge + result passthrough
Two bugs kept apps (e.g. NightRider's MIDI Pattern Editor) stuck on an empty
shell even though the host fetched the resource and called the tool correctly.
1. StrictMode double-mounted the bridge. The dev-only effect setup→cleanup→
setup built TWO host bridges/transports, double-loaded the sandbox, and
raced the app's ui/initialize handshake (it timed out, so `initialized`
never fired and buffered input/result were never flushed). AppRenderer now
defers bridge disposal to a microtask and, when the effect re-runs
synchronously with identical inputs, cancels the disposal and reuses the
live bridge — so a single bridge survives the double-invoke. The pending
input/result buffer is no longer cleared on a transient cleanup.
2. The SDK rejected the server's result. callTool() validates structuredContent
against the tool's outputSchema and throws on a mismatch — but an MCP App is
a passthrough: the host must forward whatever the server returned to the
running view (which legacy hosts render fine). InspectorClient.callTool gains
a `skipOutputValidation` option that routes through client.request() (no
wrapper validation); onOpenApp uses it.
Verified end-to-end against a real app via Playwright: the editor now renders
its grid + pattern data identically to the v1 inspector.
Tests: AppRenderer StrictMode single-bridge/no-dispose case; a get_temp_extra
test-server fixture (returns an extra structuredContent property) with
integration tests covering both the throwing default path and the
skipOutputValidation passthrough.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(apps): warn (toast) when an app's result violates its outputSchema
The skipOutputValidation passthrough keeps schema-mismatched apps rendering,
but that silently hid a real spec problem from server developers. Now the
host still renders, but surfaces the mismatch.
- InspectorClient.callTool runs the SDK's own Ajv output validation on the
skipOutputValidation path NON-fatally and returns the message as
`outputValidationError` on the ToolCallInvocation (and the
toolCallResultChange event); the result is still delivered.
- App.tsx onOpenApp shows a yellow toast when outputValidationError is set,
explaining strict MCP clients may not render the app.
Tests: integration cases asserting outputValidationError is set for a
schema-violating result (get_temp_extra) and unset for a valid one (get_temp).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(apps): summarize output-schema warning in toast + details modal
The previous toast dumped the full (very long) validation error and
overflowed. Now the toast shows a one-line summary with a "View validation
details" link that opens a modal containing the full error in a read-only,
scrollable textarea.
- New OutputValidationModal group component (+ story + tests).
- App.tsx: toast message is a summary + link; clicking it opens the modal
populated with the tool name and the full outputValidationError.
Verified via Playwright against the NightRider MIDI editor: toast is concise,
link opens the modal with the scrollable details.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(apps): address PR review — theme stability, onError, validation polish
Responds to the automated PR review on #1403.
- Theme toggle no longer rebuilds a running app's bridge. createAppBridgeFactory
reads the resolved theme from the DOM (Mantine's data-mantine-color-scheme) at
bridge-build time instead of capturing isDark in the factory deps, so the
factory identity is stable across theme flips. (#1)
- Wire AppRenderer's onError through AppsScreen -> InspectorView -> App.tsx,
surfacing bridge/runtime failures (no client, late rejection, failed
tools/call) as a red toast instead of a silent blank iframe. (#2)
- Extract validateToolOutput into a pure, unit-tested helper covering the
matching / extra-property / missing-structuredContent / error / bad-schema
cases (the no-structuredContent branch can't be produced via the SDK server,
which rejects it before sending). (#4)
- useSandboxUrl aborts its in-flight fetch on unmount (cancelled flag) to avoid
setState after unmount. (#5)
- Schema-mismatch toast: autoClose:false (details modal is one click away). (note)
- AppRenderer: add a state-machine comment block documenting the bridge
lifecycle refs. (#6)
- inspectorClient: note why the `as CallToolResult` casts are safe. (note)
Tests: AppsScreen onError surface; validateToolOutput unit suite; factory theme
read from DOM. validate + test:storybook green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs(apps): clarify theme attribute + useSandboxUrl fetch comments
Re-review follow-ups (non-blocking):
- createAppBridgeFactory: note that Mantine's data-mantine-color-scheme is only
ever light/dark (auto is resolved before paint), so no `auto` branch is needed
and matchMedia only covers an absent attribute.
- useSandboxUrl: correct the doc to say it fetches on mount AND re-fetches on
baseUrl/authToken change (idempotent), not "once on mount".
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(state): dispatch connect event after capabilities are populated
The list-state managers (tools/prompts/resources/templates) refresh on the
"connect" event and, since #1395, gate their list RPC on getCapabilities().
But connect() dispatched "connect" before fetchServerInfo() populated
this.capabilities, and refresh() runs synchronously up to its first await —
so the capability gate executed during the synchronous dispatch, read
undefined capabilities (reset to undefined on every disconnect), tripped,
and wiped every list to empty on connect. Tools, prompts, resources, and
resource templates all vanished against any server.
Move the "connect" dispatch to after the awaited fetchServerInfo() so
capabilities are in place before any manager refreshes. The unit tests
missed this because their FakeInspectorClient has capabilities populated
from construction; added an integration regression test wiring a real
ManagedToolsState on connect and asserting the tool list is non-empty
(fails without the fix).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(integration): harden connect-race regression test per review
Address the two non-blocking suggestions from the PR review:
- Replace the 100ms settle sleep with waitForEvent() on the managers'
change events, so the test is deterministic against slow CI instead of
racing a fixed timeout.
- Assert a second affected primitive (ManagedResourcesState alongside
ManagedToolsState). The fix is a single shared line, but covering two
distinct capabilities guards against a future change gating one
primitive differently.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The four list-fetching state managers fired their list RPC against every
server on connect regardless of advertised capabilities, so a server that
doesn't implement a given primitive replied -32601 "Method not found",
surfacing in the console on every connect. Mirrors the `tasks`-capability
gate added for ManagedRequestorTasksState in #1349.
- managedToolsState → gate on capabilities.tools
- managedPromptsState → gate on capabilities.prompts
- managedResourcesState → gate on capabilities.resources
- managedResourceTemplatesState → gate on capabilities.resources
(the spec defines no separate resourceTemplates capability;
resources/templates/list is part of the resources surface)
When the capability is absent each manager sets an empty list, dispatches
its change event, and returns — the right semantics for "this server
doesn't support X."
Tests: existing flow tests now construct their FakeInspectorClient with
the relevant capability so the live list path is still exercised; added a
"refresh skips listX when capability absent" and a "connect against an
X-less server doesn't fire listX" test per manager. Updated the
useManagedX peers and resourceSubscriptionsState (which drives a
ManagedResourcesState refresh) to advertise the capability too.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(auth): inject MCP_INSPECTOR_API_TOKEN into served index.html (#1378)
Reloading the web client at the bare URL (no `?MCP_INSPECTOR_API_TOKEN=…`
query string) with empty sessionStorage made every `/api/*` request 401 —
the browser had no way to recover the backend's auth token.
Embed the token into `index.html` on every page load so the browser no
longer depends on the query string surviving navigation:
- New shared helper `clients/web/server/inject-auth-token.ts` embeds
`<script>window.__INSPECTOR_API_TOKEN__ = "…"</script>` (escaped against
`</script>` injection; no-op when auth is dangerously omitted).
- Dev: the Vite plugin injects via `transformIndexHtml`.
- Prod: the Hono server injects on the `/` route.
- `App.tsx` `getAuthToken()` now reads the injected global first, then the
query string, then sessionStorage (both fallbacks preserved).
- Shared global name lives in `INSPECTOR_API_TOKEN_GLOBAL`
(`core/mcp/remote/constants.ts`).
Tests: helper unit coverage + an integration test exercising the real
prod server's `/` → `/api/*` flow (injected token authenticates; missing
token 401s). AGENTS.md documents the token-recovery order.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(auth): wire OAuth authorization-code flow into App.tsx (#1379)
OAuth-protected MCP servers could not be connected to from the v2 web
client: the core OAuth pipeline exists, but App.tsx never invoked it, so a
connect attempt 401'd and surfaced "Remote send failed (401): … Missing
Authorization header" as a toast.
Wire the two missing entry points (all core primitives already in place):
- Auto-trigger on 401: in onToggleConnection's catch, detect an upstream
401 (isUnauthorizedError) and call client.authenticate(), which runs
discovery + DCR (backend-proxied) and redirects the page to the auth
server via BrowserNavigation. The initiating server id is persisted to
sessionStorage first, since the OAuth `state` carries only mode+authId
and the full-page redirect wipes React state.
- /oauth/callback handler: a mount effect that, once `servers` hydrate,
parses the callback params, recovers the pending server, rebuilds its
InspectorClient, runs completeOAuthFlow(code) (PKCE verifier + DCR client
info survive in BrowserOAuthStorage), replaceState("/") so a reload can't
replay the single-use code, then connect(). An `error=` callback toasts
instead of retrying.
connect() already attaches the OAuth provider to the transport
(inspectorClient.ts), so once tokens land in BrowserOAuthStorage the
outbound request carries the bearer token.
Extracted the pure pieces (constants + isUnauthorizedError) to
src/utils/oauthFlow.ts with unit tests. Verified end-to-end in a real
browser against the MCP SDK demo OAuth server: Connect -> redirect ->
auto-approve -> callback -> Connected, with the access token shown in the
Connection Info modal (#1377).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(auth): persist OAuth pre-redirect Network log across the redirect (#1384)
After #1379, the Network tab showed only the post-redirect auth HTTP
(discovery re-run + POST /token); the pre-redirect discovery and the DCR
POST /register that run during authenticate() were lost when the page
navigated to the auth server.
Root causes:
1. Ordering — BrowserNavigation set `window.location.href` before the
client's `saveSession` event fired (OAuthManager calls onBeforeOAuthRedirect
*after* auth() already navigated), so the save raced the unload and was
dropped. Fix: BrowserNavigation now runs a synchronous `beforeNavigate`
hook immediately before assigning location.href; App wires it through
createWebEnvironment to flush the active fetch log to RemoteInspectorClient
Storage (keyed by the authId parsed from the auth URL) via a keepalive POST
that outlives the unload.
2. Illegal invocation — RemoteInspectorClientStorage defaulted to
`this.fetchFn = globalThis.fetch` and called `this.fetchFn(...)`, which
re-binds `this` and makes native fetch throw "Illegal invocation"
(swallowed by the catch). This silently broke *all* session save/load.
Fix: default to a wrapper that preserves the global receiver.
3. Restore race — hydrateFetchRequests replaced the list, so a load that
resolved after the resuming connect appended live entries would clobber
them. Fix: merge restored (older) entries ahead of live ones, dedupe by id.
saveSession also now uses keepalive: true.
Verified end-to-end against the MCP SDK demo OAuth server: the connected
page's Network tab shows the full handshake — pre-redirect discovery + DCR
/register plus post-redirect discovery + /token as `auth`, alongside
`transport`. Added unit tests for the beforeNavigate ordering and the
hydrate merge/dedupe.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(network): show auth response bodies with masked secrets + reveal toggle (#1386)
Auth-category Network entries showed request body/headers/status but never
the response body (rendered "(empty)"), because buildEffectiveAuthFetch
deliberately skipped capturing it to avoid surfacing access_token /
refresh_token. That hid the most useful thing to inspect when debugging
OAuth — the token exchange.
Capture auth response bodies, but mask sensitive OAuth fields by default
behind a click-to-reveal toggle so they aren't exposed at a glance during
a screen-share:
- inspectorClient: wire updateResponseBody on the auth fetcher.
- src/utils/maskSecrets.ts: maskSecretsInBody() masks access_token,
refresh_token, id_token, client_secret (case-insensitive, nested) in JSON
bodies; reports whether anything was masked. Non-JSON / secret-free bodies
pass through untouched.
- NetworkEntry BodyPreview: when a body has masked fields, render it masked
with a Reveal/Hide toggle (copy honors the shown view). Masking is a UI
concern; the raw entry is unchanged so reveal shows the real value.
access_token / refresh_token live in the post-redirect /token response,
which is never written to the session-restore files (#1384); only
pre-redirect bodies (public discovery, DCR /register) persist, so no bearer
token hits disk.
Verified end-to-end: the /token response shows masked by default
(access_token: "••••••••"), Reveal exposes the raw value, and discovery /
the public DCR /register response (no secret) render with no toggle. Added
util + component unit tests and a story play function.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(auth): inject token into prod SPA fallback + prime sessionStorage + no-store (#1378)
Addresses code-review feedback on the token-injection PR:
- Prod `server.ts` SPA deep-link fallback (e.g. `/oauth/callback`) previously
served the raw index.html off disk via serveStatic, bypassing injection — so
a bookmark/reload at a non-`/` route with empty sessionStorage would 401.
Route the SPA fallback through the same `serveIndexHtml` (inject) helper;
real static assets (paths with a dot) still serve verbatim. Dev already
injected on every HTML serve via Vite `transformIndexHtml`.
- `getAuthToken()` now persists the injected `window.__INSPECTOR_API_TOKEN__`
to sessionStorage (not just the URL-param branch), priming the backstop for
any later navigation that loses the global.
- Injected HTML responses now send `Cache-Control: no-store`, so a page
carrying a token isn't cached and served stale after a restart regenerates
the token.
Integration tests added: SPA fallback (`/oauth/callback`) carries the token,
`Cache-Control: no-store` on injected HTML, real assets served verbatim, and
unknown `/api` routes 404 rather than falling through to the HTML shell.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(auth): address #1383 review — split OAuth/connect toasts, tighten 401 match (#1379)
Code-review feedback on the OAuth-wiring PR:
- Callback effect: split completeOAuthFlow vs connect() into separate
try/catch blocks. A token-exchange failure now reads "OAuth token exchange
failed … Please try connecting again." (the single-use code is spent and the
URL was cleared, so a reload can't retry); a post-OAuth connect failure reads
"Failed to connect" since OAuth actually succeeded and re-clicking Connect
reuses the persisted tokens.
- isUnauthorizedError: anchor the message fallback on the transport's
`failed …(401)` wording instead of a bare `(401)`, so an unrelated `(401)`
spliced into an error message can't trip the OAuth flow. Added a test.
- Documented that clearing the pending id + URL before the server lookup is
intentional (deleted/renamed server mid-flow → require a fresh Connect).
Also merges the squash-merged #1382 base from v2/main.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs(auth): address #1385 review — clarify double-save + keepalive cap + test fetch default (#1384)
Code-review feedback on the OAuth Network-log persistence PR:
- Documented the double-save: `FetchRequestLogState`'s `saveSession` listener
is the backstop; `BrowserNavigation`'s `beforeNavigate` hook is the primary
flush for the redirect case. Notes the listener may lose the navigation race
and is harmless when it duplicates (last-writer-wins, identical payload).
- Reworded the keepalive comment in `RemoteInspectorClientStorage.saveSession`:
the 64KB cap is general (the method is also reachable from the listener with
the full session log), so a long session could exceed it and drop silently —
acceptable since the persisted log is best-effort, not load-bearing.
- Added a regression test that constructs `RemoteInspectorClientStorage`
without a `fetchFn`, stubs `globalThis.fetch`, and asserts the default
wrapper calls it (locks in the "Illegal invocation" fix, which callers
otherwise swallow).
Optional items (logger.warn on swallowed save errors; setupClientForServer dep
churn) acknowledged on the PR, not changed.
Also merges the squash-merged #1383 base from v2/main.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(network): address #1387 review — form-encoded masking, cleaner flag, a11y (#1386)
Code-review feedback on the auth-response-body masking PR:
- Extended masking to form-urlencoded bodies (finding #1): the token *request*
is `application/x-www-form-urlencoded` and carries `code` / `code_verifier` /
`client_secret` / `refresh_token`. `maskSecretsInBody` now masks those in
form bodies too (preserving formatting; placeholder not percent-encoded).
`code`/`code_verifier`/`client_assertion` are form-only sensitive keys —
deliberately NOT masked in JSON, where `code` is usually an error/status code.
- Replaced the double-stringify `hasSecrets` heuristic with an explicit
masked-flag propagated out of `maskNode` (finding #2) — robust if the
transform ever grows non-identity behavior, and one fewer serialization.
- Reset reveal state on body change via `key={body}` remount instead of a
setState-in-effect (finding #3; avoids the cascading-render lint rule).
- a11y (finding #4): `aria-label` on the Reveal/Hide button and `aria-live`
on the hidden/revealed status text.
- Security tripwire comment near the `saveSession` listener (finding #6):
captured auth bodies are unmasked at source (masking is UI-only), so any new
post-token-exchange persistence path must redact first.
Tests: form-encoded masking (token + refresh requests), JSON `code` NOT masked,
non-object JSON pass-through, default-fetch wrapper. Story play + NetworkEntry
tests updated for the new aria-labels and the now-masked form request body.
Finding #5 (pretty-print asymmetry) is already handled by ContentViewer's JSON
formatting; #4-aria and #6 are the doc/a11y bits.
Also merges the squash-merged #1385 base from v2/main.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(network): address #1387 third-pass review — content-type masking, wholesale mask, dedup (#1386)
Third-pass code-review feedback on the auth-body masking PR:
- maskSecretsInBody now takes the body's content-type (finding #1): `*json*`
→ JSON masking, form-urlencoded → form masking, any other known type
(HTML/plaintext/XML) → no masking; absent/unknown → sniff as before. Removes
the implicit "non-JSON ⇒ form" guess for error pages etc. NetworkEntry passes
the request/response `content-type` header through to BodyPreview.
- Mask any non-empty value under a sensitive key wholesale (finding #2):
a non-standard object/array value under e.g. `access_token` is replaced
rather than recursed into, so it can't leak. Empty-string still not flagged.
- Extracted `isSensitiveKey(set, key)` to dedupe the JSON/form key checks
(finding #3).
- Reworded `MaskResult.masked` doc to cover the form (non-pretty-printed) case
(finding #6).
Tests: non-string-value-under-sensitive-key wholesale mask; repeated form
param; empty form value not flagged; content-type honored (JSON type skips
form masking, text/plain skips masking, explicit form type masks).
Finding #5 (story re-hide) skipped per reviewer — unit test already covers
re-mask. Finding #4 (form edge cases) added.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(network): address #1387 fourth-pass nits — content-type contract, key, form test (#1386)
Fourth-pass review (LGTM) — minor, non-blocking items:
- Documented the content-type matching contract on `maskSecretsInBody`:
substring match (`*json*` / `*x-www-form-urlencoded*`), and we trust the
wire's own label (a mislabeled body renders raw — acceptable for the
screen-share threat model) (finding #1).
- Key `<BodyPreview>` by content-type + body (not body alone) so a header-only
change would also reset the reveal state (finding #3; not reachable today).
- Storybook `AuthSuccess`: assert revealed count `>=` reveal-button count to
mirror the `hidden >= 2` check, so adding a non-masked body later can't drift
the assertion silently (finding #4).
- Added a `NetworkEntry` unit test for form-encoded request-body masking
(code/code_verifier) — previously exercised only via the Storybook play.
Finding #2 (silent decodeURIComponent fallback on malformed keys) acknowledged
as a known quirk outside the threat model; no change.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* perf(network): address #1387 fifth-pass — memoize masking, mask DCR mgmt token (#1386)
Fifth-pass review (LGTM):
- Memoized `maskSecretsInBody` in `BodyPreview` (recommended item #1): a
Reveal/Hide click no longer re-parses + re-walks the body; cost is once per
mount, and the `key` remount re-runs it on body/content-type change. Guarded
so a too-large body is never parsed (the hook runs unconditionally, with the
size check inside the memo).
- Added `registration_access_token` (RFC 7592 DCR management credential, same
bearer class) to the masked key set, and a confidential-client `/register`
response fixture test (#5) asserting client_secret + registration_access_token
masked, client_id/metadata visible.
- Tightened the `isMaskableValue` doc to state the exact contract: non-null,
non-empty-string values are masked wholesale (#3).
Items #2 (transport parse cost — capped by the memo) and #4 (test-only
MASK_PLACEHOLDER export) left as-is per the review.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs(network): clarify body-update emits list event only (#1387 sixth pass) (#1386)
Sixth-pass review (LGTM, all nits). Added a comment on
`onFetchRequestBodyUpdate` noting it re-emits only `fetchRequestsChange`, not
the per-entry `fetchRequest` event — list-reading consumers pick up the body
on the next render; a future per-entry subscriber would need its own event.
Other items left as-is per the review: full-string `BodyPreview` key (#1,
micro-perf only if profiled), NaN under a sensitive key (#3, consistent with
the mask-non-null-non-empty contract), `&`-only form separator (#4, OAuth
never emits `;`).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(auth): inject MCP_INSPECTOR_API_TOKEN into served index.html (#1378)
Reloading the web client at the bare URL (no `?MCP_INSPECTOR_API_TOKEN=…`
query string) with empty sessionStorage made every `/api/*` request 401 —
the browser had no way to recover the backend's auth token.
Embed the token into `index.html` on every page load so the browser no
longer depends on the query string surviving navigation:
- New shared helper `clients/web/server/inject-auth-token.ts` embeds
`<script>window.__INSPECTOR_API_TOKEN__ = "…"</script>` (escaped against
`</script>` injection; no-op when auth is dangerously omitted).
- Dev: the Vite plugin injects via `transformIndexHtml`.
- Prod: the Hono server injects on the `/` route.
- `App.tsx` `getAuthToken()` now reads the injected global first, then the
query string, then sessionStorage (both fallbacks preserved).
- Shared global name lives in `INSPECTOR_API_TOKEN_GLOBAL`
(`core/mcp/remote/constants.ts`).
Tests: helper unit coverage + an integration test exercising the real
prod server's `/` → `/api/*` flow (injected token authenticates; missing
token 401s). AGENTS.md documents the token-recovery order.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(auth): wire OAuth authorization-code flow into App.tsx (#1379)
OAuth-protected MCP servers could not be connected to from the v2 web
client: the core OAuth pipeline exists, but App.tsx never invoked it, so a
connect attempt 401'd and surfaced "Remote send failed (401): … Missing
Authorization header" as a toast.
Wire the two missing entry points (all core primitives already in place):
- Auto-trigger on 401: in onToggleConnection's catch, detect an upstream
401 (isUnauthorizedError) and call client.authenticate(), which runs
discovery + DCR (backend-proxied) and redirects the page to the auth
server via BrowserNavigation. The initiating server id is persisted to
sessionStorage first, since the OAuth `state` carries only mode+authId
and the full-page redirect wipes React state.
- /oauth/callback handler: a mount effect that, once `servers` hydrate,
parses the callback params, recovers the pending server, rebuilds its
InspectorClient, runs completeOAuthFlow(code) (PKCE verifier + DCR client
info survive in BrowserOAuthStorage), replaceState("/") so a reload can't
replay the single-use code, then connect(). An `error=` callback toasts
instead of retrying.
connect() already attaches the OAuth provider to the transport
(inspectorClient.ts), so once tokens land in BrowserOAuthStorage the
outbound request carries the bearer token.
Extracted the pure pieces (constants + isUnauthorizedError) to
src/utils/oauthFlow.ts with unit tests. Verified end-to-end in a real
browser against the MCP SDK demo OAuth server: Connect -> redirect ->
auto-approve -> callback -> Connected, with the access token shown in the
Connection Info modal (#1377).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(auth): persist OAuth pre-redirect Network log across the redirect (#1384)
After #1379, the Network tab showed only the post-redirect auth HTTP
(discovery re-run + POST /token); the pre-redirect discovery and the DCR
POST /register that run during authenticate() were lost when the page
navigated to the auth server.
Root causes:
1. Ordering — BrowserNavigation set `window.location.href` before the
client's `saveSession` event fired (OAuthManager calls onBeforeOAuthRedirect
*after* auth() already navigated), so the save raced the unload and was
dropped. Fix: BrowserNavigation now runs a synchronous `beforeNavigate`
hook immediately before assigning location.href; App wires it through
createWebEnvironment to flush the active fetch log to RemoteInspectorClient
Storage (keyed by the authId parsed from the auth URL) via a keepalive POST
that outlives the unload.
2. Illegal invocation — RemoteInspectorClientStorage defaulted to
`this.fetchFn = globalThis.fetch` and called `this.fetchFn(...)`, which
re-binds `this` and makes native fetch throw "Illegal invocation"
(swallowed by the catch). This silently broke *all* session save/load.
Fix: default to a wrapper that preserves the global receiver.
3. Restore race — hydrateFetchRequests replaced the list, so a load that
resolved after the resuming connect appended live entries would clobber
them. Fix: merge restored (older) entries ahead of live ones, dedupe by id.
saveSession also now uses keepalive: true.
Verified end-to-end against the MCP SDK demo OAuth server: the connected
page's Network tab shows the full handshake — pre-redirect discovery + DCR
/register plus post-redirect discovery + /token as `auth`, alongside
`transport`. Added unit tests for the beforeNavigate ordering and the
hydrate merge/dedupe.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(auth): inject token into prod SPA fallback + prime sessionStorage + no-store (#1378)
Addresses code-review feedback on the token-injection PR:
- Prod `server.ts` SPA deep-link fallback (e.g. `/oauth/callback`) previously
served the raw index.html off disk via serveStatic, bypassing injection — so
a bookmark/reload at a non-`/` route with empty sessionStorage would 401.
Route the SPA fallback through the same `serveIndexHtml` (inject) helper;
real static assets (paths with a dot) still serve verbatim. Dev already
injected on every HTML serve via Vite `transformIndexHtml`.
- `getAuthToken()` now persists the injected `window.__INSPECTOR_API_TOKEN__`
to sessionStorage (not just the URL-param branch), priming the backstop for
any later navigation that loses the global.
- Injected HTML responses now send `Cache-Control: no-store`, so a page
carrying a token isn't cached and served stale after a restart regenerates
the token.
Integration tests added: SPA fallback (`/oauth/callback`) carries the token,
`Cache-Control: no-store` on injected HTML, real assets served verbatim, and
unknown `/api` routes 404 rather than falling through to the HTML shell.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(auth): address #1383 review — split OAuth/connect toasts, tighten 401 match (#1379)
Code-review feedback on the OAuth-wiring PR:
- Callback effect: split completeOAuthFlow vs connect() into separate
try/catch blocks. A token-exchange failure now reads "OAuth token exchange
failed … Please try connecting again." (the single-use code is spent and the
URL was cleared, so a reload can't retry); a post-OAuth connect failure reads
"Failed to connect" since OAuth actually succeeded and re-clicking Connect
reuses the persisted tokens.
- isUnauthorizedError: anchor the message fallback on the transport's
`failed …(401)` wording instead of a bare `(401)`, so an unrelated `(401)`
spliced into an error message can't trip the OAuth flow. Added a test.
- Documented that clearing the pending id + URL before the server lookup is
intentional (deleted/renamed server mid-flow → require a fresh Connect).
Also merges the squash-merged #1382 base from v2/main.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs(auth): address #1385 review — clarify double-save + keepalive cap + test fetch default (#1384)
Code-review feedback on the OAuth Network-log persistence PR:
- Documented the double-save: `FetchRequestLogState`'s `saveSession` listener
is the backstop; `BrowserNavigation`'s `beforeNavigate` hook is the primary
flush for the redirect case. Notes the listener may lose the navigation race
and is harmless when it duplicates (last-writer-wins, identical payload).
- Reworded the keepalive comment in `RemoteInspectorClientStorage.saveSession`:
the 64KB cap is general (the method is also reachable from the listener with
the full session log), so a long session could exceed it and drop silently —
acceptable since the persisted log is best-effort, not load-bearing.
- Added a regression test that constructs `RemoteInspectorClientStorage`
without a `fetchFn`, stubs `globalThis.fetch`, and asserts the default
wrapper calls it (locks in the "Illegal invocation" fix, which callers
otherwise swallow).
Optional items (logger.warn on swallowed save errors; setupClientForServer dep
churn) acknowledged on the PR, not changed.
Also merges the squash-merged #1383 base from v2/main.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
cliffhall
Bob Dickinson