Rename the web client's History tab and its supporting code to Protocol
so the feature reads as the MCP "Protocol" view.
- Tab renders as "Protocol"; INSPECTOR_TAB_IDS, InspectorView tab
constants/MonitorTab, TAB_UI_REGISTRY key, and navigation.ts tab id all
updated to Protocol.
- Rename components + folders: HistoryScreen -> ProtocolScreen,
HistoryListPanel -> ProtocolListPanel, HistoryEntry -> ProtocolEntry,
HistoryControls -> ProtocolControls, historyUtils -> protocolUtils
(plus their tests and stories, and exported types).
- Rename App/InspectorView plumbing: historyUi, pinnedHistoryIds,
on*History* handlers/props, replayHistoryRequest,
isReplayableHistoryMethod, sort/compact scopes, etc. -> Protocol names.
- Export filenames use the "protocol" slug (buildExportFilename +
ExportKind).
Preserved (per issue #1623 decisions):
- The in-panel section discriminator keeps its "pinned" | "history"
values, and the sub-sections stay labeled Pinned / History.
- window.history and core message-log/replay concepts are untouched.
npm run ci passes (format, lint, build, unit tests, >=90 per-file
coverage gate, smoke, Storybook).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BrihGWcrM9JGRyu41nzZYw
Re-review follow-up (#1622):
- Observation A: a returning user whose last-pinned monitor tab was
History would land on the (empty) History tab on a stdio failure instead
of the Console diagnostic. Root cause: History is *always* empty on a
connect failure (the message log clears on the error transition), yet it
was pushed unconditionally, so it could precede/hide the real diagnostic.
Fix: content-gate History like Console/Network — only offer it when it
has entries. On a fresh failure the diagnostic (Console/Network) is the
only tab, so it always leads, without touching tab-switching or adding
state. This also resolves the earlier "opens onto an empty column" note:
when nothing has been captured yet, monitorAvailable is empty and the
column stays closed, opening reactively once stderr/fetch content lands.
Tests: stdio/HTTP failure now assert the diagnostic leads and History is
absent; added "leads with Console even when stored tab was History",
"keeps the column closed until a diagnostic has content", and "offers
History when it has captured content". Verified live (time server →
Console-only column). Full `npm run ci` passes.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BrihGWcrM9JGRyu41nzZYw
Review feedback on #1622:
- #1 (medium): `connectionStatus === "error"` also covers a mid-session
crash of a previously-connected server, not just a failed connect
attempt. Scope the auto-open + failure tab set to connect attempts only:
the auto-open effect's error arm now requires the previous status to not
be `connected` (connecting/disconnected → error), and `failed` (render)
requires `erroredServerId`, which the parent sets only for connect-
attempt failures. A mid-session crash now closes the column like any
teardown instead of reopening a user-closed column / dropping the live
Logs tab. This also aligns the column behavior with the red border.
- #3 (minor): extract the column slide duration into a shared
`monitorColumnAnimation.ts` (MONITOR_COLUMN_ANIM_MS +
FAILED_CARD_SCROLL_DELAY_MS) so InspectorView's Transition and
ServerCard's scroll delay can't drift.
- #4 (nit): composite key for ConsoleScreen entry rows instead of the
array index.
Added a mid-session-crash test (connected → error with no erroredServerId
closes the column). Verified live: a connect failure still opens the
column to Console. Full `npm run ci` passes.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BrihGWcrM9JGRyu41nzZYw
Opening the monitoring column on a failed connection reflows the server
grid to fewer/narrower columns, which can push the failed card below the
fold. Scroll it into view so the red-bordered card is visible without
hunting for it.
- ServerCard scrolls itself into view on the disconnected→errored
transition (mirrors the freshly-added highlight scroll), deferred
~320ms so the column open + grid reflow settle first — otherwise
scrollIntoView targets the card's pre-reflow position. Guarded by a ref
so it fires once per failure and won't fight a user who scrolls away.
Verified live: the failed `time` card scrolls fully into view after the
column opens. Full `npm run ci` passes.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BrihGWcrM9JGRyu41nzZYw
Flags the server whose last connection attempt failed with a red card
border, cleared when another server is connected or a new connection is
attempted — a persistent, at-a-glance companion to the failure toast and
the auto-opened monitoring column.
- App tracks `failedServerId`: set in onToggleConnection's non-auth
handshake catch, cleared at the start of any new connect attempt. It is
deliberately kept out of resetSessionScopedUiState, since the failure's
own `disconnect` event would otherwise clear it the same tick (that same
event clears `activeServer`, which is why the flag is tracked separately).
- New `errored` Card theme variant (red border via a new
`--inspector-error-border` token), mirroring the `highlighted` green
border. Precedence: disabled (dimmed) > errored > highlighted.
- Plumbed `erroredServerId` through InspectorView → ServerListScreen →
ServerCard (`errored` prop).
Full `npm run ci` passes (validate + coverage + smoke + storybook).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BrihGWcrM9JGRyu41nzZYw
The InspectorView story args predate the new required Console props, so the
"Connection Error" play story crashed rendering (`stderrLogs` undefined →
`.length` throw) — caught by the Storybook step of `npm run ci`, which I
should have run before the previous push.
- Add `stderrLogs` / `consoleUi` / `onConsoleUiChange` / `onClearConsole` /
`onExportConsole` to the story args, with a `demoStderr` fixture.
- Make the Connection Error story a stdio-failure narrative: it now opens
the column to Console showing the captured startup error, with `network:
[]` so Network isn't offered.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BrihGWcrM9JGRyu41nzZYw
A failed stdio launch had no useful diagnostic surface: the message log
(History) is cleared on the error transition and there are no HTTP
requests (Network), so the auto-opened monitoring column was empty. The
process's stderr — which explains the failure (e.g. "No module named
mcp_server_time") — was captured in StderrLogState but never displayed.
- New ConsoleScreen renders the captured stderr stream (search / sort /
clear / export; embeddable in the monitoring column), wired through App
via the existing useStderrLog hook + StderrLogState.
- Console is a pinnable monitor tab like Logs/History/Network, gated to
stdio servers (HTTP has no child process). It lives in the header menu
when the column isn't pinned, and moves into the column when it is.
- The failure column now picks its tabs by which diagnostic actually
captured content (stderr → Console, fetch requests → Network), not by
the active server's transport: a connect failure fires the client
`disconnect` event, which clears `activeServer`, so transport can't be
read back there. The memo re-runs as stderr/fetch entries stream in, so
the diagnostic tab appears as soon as the failing process prints.
Verified live against a failing `python -m mcp_server_time` stdio server:
the column auto-opens to Console showing the real ModuleNotFoundError.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BrihGWcrM9JGRyu41nzZYw
On a connection failure the user was left without an obvious way to see
why. Open the monitoring column automatically on the transition into the
`error` status (mirroring the connect-success behavior from #1616) so the
diagnostics surface right when they're needed.
- Broaden the auto-open effect to fire on the transition into `error` as
well as `connected`, tracked via a general previous-status ref so it
fires once per transition and never on a mount that starts errored.
- A failed connect never negotiated capabilities, so the failure column
offers History (the local client-side message log) and — for non-stdio
transports — Network (HTTP requests), the entries that explain the
failure. Logs is omitted (it needs the server `logging` capability).
- `effectivePinned` now allows the column while `error`, not only while
`connected`; it still closes on a clean disconnect or narrowing.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BrihGWcrM9JGRyu41nzZYw
The root ci:integration alias became dead after ci switched to run
coverage (which runs web's unit+integration projects under v8). Its
only occurrence was its own definition. clients/web test:integration
remains available for direct manual runs.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BrihGWcrM9JGRyu41nzZYw
The local `npm run ci` was not at parity with `.github/workflows/main.yml`.
CI runs `validate → coverage → smoke → test:storybook`, but the root `ci`
script ran `validate → ci:integration → smoke → ci:storybook`, omitting
`npm run coverage` — the per-file ≥90 (lines/statements/functions/branches)
gate that CI enforces. So a push could pass every documented local pre-push
command and still fail CI on the coverage gate (this exact gap broke PR #1601
on a HistoryEntry.tsx function-coverage regression).
Two changes:
1. Root `package.json` `ci` script now chains `npm run coverage` in place of
`ci:integration`. Because web's `coverage:web` (`test:coverage`) already
runs the integration project under instrumentation, `coverage` supersedes
the old integration-only step and makes `ci` a genuine superset of GitHub
CI. The `ci:integration` / `ci:storybook` script definitions are left in
place for any other references.
2. AGENTS.md: rename the pre-push section to "Mandatory pre-push gate", make
running the coverage gate a hard non-optional step (remove the "or at
minimum `npm run validate`" loophole and the soft "optionally also run
coverage" phrasing), and correct the parity claim so `npm run ci` is
described as `validate → coverage → smoke → storybook`.
Closes#1619
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BrihGWcrM9JGRyu41nzZYw
Adds a test that clicks Expand/Collapse in the embedded (compact)
layout, exercising the ExpandToggle onToggle callback that was
uncovered (line 193). Restores HistoryEntry.tsx to 100% coverage,
fixing the CI per-file function-coverage gate failure (was 85.71%).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BrihGWcrM9JGRyu41nzZYw
Auto-open the monitoring column when a connection is established. Gated on the
disconnected → connected transition (a ref-tracked edge), so it fires on an
actual connect rather than on every render while connected or on a mount that
starts already-connected (which would fight a user who closed it). The column
still only appears when wide and a monitor tab is available.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
At the 320px column minimum the full ISO timestamp ate most of line 1. Use an
HH:MM:SS (UTC) form in the embedded HistoryEntry/NetworkEntry headers only; the
full-size layouts keep the full ISO string.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
In the pinned monitoring column (embedded), the single-line entry headers wrap
badly. Add compact two-line variants gated on `embedded`:
- HistoryEntry: line 1 = time + client/server badge + elapsed + status; line 2 =
method (and target, truncated) on the left with the controls on the right,
where Replay becomes an icon (new ReplayButton element) next to Pin/Expand.
- NetworkEntry: line 1 = time + HTTP method + category + elapsed + status; line 2
= the URL in a horizontal ScrollArea (new Text `nowrap` variant) with the
expand toggle on the right.
The full-size (non-embedded) layouts are unchanged. `embedded` is threaded from
the History/Network panels into the entries.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Move the Client/Server direction badge to the right of the timestamp so the
monospace timestamp is leftmost, matching the NetworkEntry header layout.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Fix the resize handle disappearing: the animation wrapper is a Mantine Group
(default align: center), and the full-height handle has no intrinsic height, so
it collapsed to 0. Set align: stretch on the wrapper.
- Closing the column now leaves the primary on the screen currently selected in
the header (commits the active tab first), instead of swapping in the column's
screen.
- Keep the primary screen's standard ScreenStage transition; only the column
(handle + column) slides open/closed from the right with a fade.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Add a search box to MonitoringControls (right of the tabs). It's a single
shared search kept across column tab switches, filtering whichever screen is
shown. Embedded panels apply this text but still ignore the sidebar-only
filters (levels/categories/directions/method), which have no control there.
- Close button now uses the "collapse sidebar" icon (TbLayoutSidebarRightCollapse),
the mirror of the pin button's expand icon.
- Closing the column carries its selection to the primary view: the screen the
user was watching becomes the active header tab (rather than snapping back to
the pre-pin selection).
- Animate the column open/close: the handle + column slide in from the right and
fade (reversing on close) via a Mantine Transition.
- HistoryListPanel: drop the redundant "History (N)" header when there are no
pinned requests (lone unpinned section renders headerless).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- ResizeHandle: clear the `resizing-col` body class on unmount so a drag
interrupted by the handle unmounting (narrow past the breakpoint / disconnect)
can't leave page-wide user-select:none / col-resize applied.
- Embedded monitor column now shows the full unfiltered stream instead of
silently mirroring the full-size screen's live filter (the column has no
filter control to explain an empty/partial list).
- MonitoringControls: give the SegmentedControl an aria-label ("Monitoring
screen") so the tab group has an accessible name.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Render PinColumnButton as a subtle size-sm Button with a size-20 icon, the same
style as the panel's ListToggle, instead of a default ActionIcon.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
On the History and Network panels, place the ListToggle (expand/collapse)
after Export and before the pin-as-column button. (Logs has no such toggle.)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Place the "pin as column" icon after Clear/Export (rightmost) on the Logs,
History, and Network panel toolbars instead of at the front.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add the ability to "pin" the Logging, History, and Network screens into a
second column on the right of the InspectorView, so a primary screen and a
monitor screen are visible at once (e.g. watch Logs while calling a Tool).
- A "pin as column" button on the Logs/History/Network panel toolbars opens
the column (shown only when wide + connected).
- The column hosts a MonitoringControls tab row (available monitor tabs +
close) over the selected embedded screen. Pinning removes the monitor group
from the header tab bar; closing (x) returns it.
- The divider between the primary area and the column is draggable (pointer +
keyboard); width is clamped and persisted.
- The split collapses to one column below ~1040px (the ServerListScreen
single-card breakpoint) and while disconnected, without discarding the pin
preference — it reopens when wide + connected again.
State (pinned/tab/width) is view-local and persisted via the existing
useLocalStorage pattern, so no new props thread through App and ViewHeader is
unchanged (it just receives a filtered tab list). Screens/panels gain an
`embedded` mode (fills the parent height, drops the 340px filter sidebar) via a
shared EmbeddableScrollArea element.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
`tsc -b` (run in CI's validate step, but not by vitest locally) rejected
`ok.mock.calls[0][1].body` — under the project's strict indexing the
untyped mock's call tuple is `[]`, so `[0]` is possibly-undefined and
`[1]` is out of range. Capture the request body inside a `vi.fn<typeof
fetch>` implementation instead of indexing into `mock.calls`.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Extend the Node-free guard beyond oauth-persist.ts to every module
reachable from the browser's RemoteOAuthStorage (storage-remote,
oauth-storage, store, oauth-persist, store-serialize), so a forbidden
store-io/node:/atomically import added one hop away is caught too — not
just a direct import in oauth-persist.ts. Addresses the non-blocking
hardening note on Claude review of #1612.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The refactor routed the browser's RemoteOAuthStorage through
`oauth-persist.ts`, which value-imported the Node-only `store-io.js`
(`atomically` -> `stubborn-fs` -> `node:process.getuid`, plus
`node:fs/promises`). That dragged Node built-ins into the browser bundle
and the web app crashed to a blank page at runtime — a regression from
v2/main, where browser OAuth storage used a Zustand JSON adapter with no
Node imports. Unit/integration tests run in node/happy-dom and
`smoke:web` only checks the served HTML, so none of them execute the app
in a real browser and the break slipped through.
Fix, following the existing `store-id.ts` precedent:
- Split the pure JSON (de)serializers into Node-free
`core/storage/store-serialize.ts`; `store-io.ts` re-exports them so
existing importers are unaffected.
- Move the Node-only `createFileOAuthPersistBackend` into
`core/auth/node/oauth-persist-file.ts`; update `storage-node.ts` and
the storage adapters test to import from there.
- `oauth-persist.ts` now imports only the Node-free serializers, so the
browser bundle no longer pulls in `node:fs`/`atomically`.
Tests:
- oauth-persist-browser-safe.test.ts: source-level guard that
`oauth-persist.ts` never imports store-io/node:/atomically.
- Direct unit tests for the isomorphic remote + session backends
(restoring oauth-persist.ts coverage after the file backend moved out).
- store.test.ts: OAuthMemoryStore unit coverage incl. replace() fallbacks.
- storage.test.ts: clearAllOAuthClientState null-snapshot (no file) path.
Verified end-to-end in a real browser (Playwright): DCR OAuth flow
completes, tokens persist via the remote store, and after a full page
reload the server reconnects with no re-authentication.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Document the contract the remote read() path leans on — the server's
missing-file response is `c.json({}, 200)`, and the backend now passes
that empty object straight to parseOAuthPersistBlob. Addresses the
optional note on Claude review of #1612.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
parseOAuthPersistBlob already returns null for {} (the server's
missing-file response) and for a literal null body, so the
Object.keys(store).length === 0 guard was redundant — and would throw a
TypeError on a null body before parseOAuthPersistBlob could handle it.
Pass the parsed res.json() straight through.
Addresses Claude review item #1 on #1612.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Review follow-ups on the RemoteOAuthStorage web migration:
- useEmaIdpLoginState.logout(): add a .catch to the now-async
clearEmaIdpSession call so a storage failure no longer surfaces as an
unhandled rejection; leave loginState unchanged on failure (the session
was not actually cleared). Add a test asserting no unhandled rejection.
- createOAuthStore: attach a no-op catch to the eager persistLoadPromise
so a hydration read that fails before any load() awaits it does not
become an unhandled rejection; real awaiters still observe the rejection
via the shared reference.
- storage-remote.test: replace `as unknown as typeof fetch` double-casts
with the typed vi.fn<typeof fetch>() pattern the same PR established.
- remoteOAuthStorage: correct the doc comment to describe the cache-of-one
accurately instead of implying a keyed map.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Serialize persist() in OAuthStorageBase so overlapping remote POSTs cannot
reorder and clobber state. Regenerate launcher package-lock.json to drop
stale zustand entries. Let parseOAuthPersistBlob accept parsed objects
from res.json() without a redundant JSON round-trip.
Addresses the @claude review of #1611: the event registry is now keyed by
(client, fn) rather than event name alone, and a new `fireClientEventFor(client,
…)` helper fires a single client's listeners. The "ignores auth lifecycle events
from a non-selected server" test now fires ambient/recovered/oauthError for ONLY
the non-selected client and asserts no message appears — so removing the
`selectedServerRef.current !== serverName` guard would fail the test — then fires
for the selected client to confirm it does act. Coverage of App.tsx is unchanged
(92.23 / 90.20 / 92.85 / 92.68); existing fire-all `fireClientEvent` behavior is
preserved.
Part of #1610.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3fTN8H3R8YV4yUGvZjYnX
Bring clients/tui/src/App.tsx from 79% to ≥90 on all four dimensions
(92.23 / 90.20 / 92.85 / 92.68) by exercising the mid-session / step-up auth
machinery that the existing suite didn't reach:
- runOAuthAuthentication result branches (success / already_authorized /
insufficient_scope / unsupported), driven deterministically via a new opt-in
runRunnerInteractiveOAuth override (defaults to the real runner so the
existing callback-driven OAuth tests are unchanged).
- The per-client auth-lifecycle listeners (authChallengeAmbient / Recovered /
Interactive / oauthError), fired through a new client-event registry on the
FakeClient, including the non-selected-server early-return guards.
- handleAuthRecoveryRequired step-up-confirm and reauth paths, handleConnect's
OAuth-on-401 branches (AuthRecoveryRequired + EmaClientNotConfigured thrown
mid-flow), handleClearOAuth's connected/disconnect path, and the AuthTab
onAuthorizeStepUp handler (EMA satisfied/interactive/failed/throw + standard
non-EMA authorize, cancel).
Part of #1610.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3fTN8H3R8YV4yUGvZjYnX
Bring clients/tui/src/components/AuthTab.tsx to ≥90 on all four dimensions
(now 97.43/90.81/100/100) by exercising the previously-untested step-up
keypress paths in the `useInput` handler: up-arrow navigation back to the
"authorize" choice, the `a` (authorize) and `c` (cancel) shortcuts, and the
fall-through for an unrelated key while the prompt is pending.
Part of #1610.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3fTN8H3R8YV4yUGvZjYnX
Bring clients/cli/src/cliOAuth.ts from 75% to ≥90% on all four coverage
dimensions (now 100/96.42/100/100) by exercising the previously-untested
interactive step-up and auth-recovery paths:
- confirmStepUpFromStdin (the default stdin confirmer): drive it via the
default `confirmStepUp` parameter of handleCliAuthRecoveryRequired with
node:readline/promises mocked — covers the y / "yes" (trimmed,
case-insensitive) accept paths and the n decline path (which throws).
- connectInspectorWithOAuth's AuthRecoveryRequiredError branch: both the
storage-already-satisfies resume and the interactive-recovery sub-branches,
plus the isUnauthorizedError branch (with a rejecting disconnect to hit the
`.catch(() => {})` guard), the non-OAuth-capable-config rethrow, and the
non-OAuth-error rethrow.
Part of #1610 — v2/main mid-session-auth code below the per-file coverage
gate that v2/main's CI does not run but the Wave-1 rollup (#1550) does.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3fTN8H3R8YV4yUGvZjYnX
- transport.ts: update the SSE-branch comment to say the default path uses
fetchWithOptionalAuthIntercept (proxy-aware baseFetch + optional auth-challenge
intercept), not the bare proxy-aware baseFetch.
- clients/cli/README.md: fix the §5 mid-session validation link to use the
../../specification/ prefix like the sibling links in the same sentence
(previously resolved to a nonexistent clients/cli/ path).
Both flagged by the @claude review of PR #1601; comment/doc-only, no runtime change.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3fTN8H3R8YV4yUGvZjYnX
