fix: make the proxy deadline per-request, not unconditional (#2319 round 13)

The server-side timer added in round 9 applied to every /api/fetch
request. But that route carries MCP traffic as well as OAuth work, so a
Streamable HTTP tool call that legitimately withholds its response
headers past the budget was aborted there — reintroducing on the
backend exactly the regression exemptMcpEndpoint prevents on the
client, and reporting it as an OAuth timeout besides.

The caller now decides. `withOAuthRequestTimeout` stamps its budget on
the RequestInit it hands down, `createRemoteFetch` reads it back and
puts it in the /api/fetch envelope, and the route applies a timer only
when one is present — so an exempt MCP request, which never gets an
init of its own, is bounded by nothing.

The carrier is a WeakMap keyed on the init rather than a header:
serializeRequest copies request headers verbatim into the payload and
the route re-sends them to the authorization server, so a marker header
would leak to a third party. It needs no cast onto RequestInit and is
collected with the object.

The `fetchTimeoutMs` option from round 9 is gone — the deadline now
lives in exactly one place, the client wrapper, which is also what
makes the 504 marker honest: only a request the OAuth wrapper bounded
can produce one.

Ten tests: the stamping and its rounding, nothing stamped on an exempt
call or an init that never went through the wrapper, the envelope
carrying the budget (and not a header), its omission on both exempt and
direct calls, and at the route the 504 marker, no deadline without one,
and a non-numeric deadline ignored. The no-deadline case is verified by
mutation: an unconditional timer makes it fail.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YFnM6RqB4nXqNysgUYmjmq
Signed-off-by: cliffhall <cliff@futurescale.com>

clients/web/src/test/core/auth/requestTimeout.test.ts

@@ -25,6 +25,7 @@ import { describe, it, expect, vi, afterEach } from "vitest";
 import {
   DEFAULT_OAUTH_REQUEST_TIMEOUT_MS,
   OAuthRequestTimeoutError,
+  deadlineForRequestInit,
   exemptMcpEndpoint,
   withOAuthRequestTimeout,
 } from "@inspector/core/auth/requestTimeout.js";
@@ -414,6 +415,46 @@ describe("withOAuthRequestTimeout", () => {
     });
   });
 
+  describe("the deadline stamped on the init (#2319 proxy hop)", () => {
+    // How the budget reaches `/api/fetch` without travelling upstream: a header
+    // would be copied verbatim into the payload and re-sent to the
+    // authorization server, so the carrier is a WeakMap keyed on the init.
+    it("stamps the budget on the init it hands down", async () => {
+      const inner = vi.fn<typeof fetch>().mockResolvedValue(new Response("{}"));
+      const wrapped = withOAuthRequestTimeout(inner, 1234);
+
+      await wrapped(URL_UNDER_TEST);
+
+      expect(deadlineForRequestInit(inner.mock.calls[0][1])).toBe(1234);
+    });
+
+    it("stamps the rounded budget, matching what is reported", async () => {
+      const inner = vi.fn<typeof fetch>().mockResolvedValue(new Response("{}"));
+      const wrapped = withOAuthRequestTimeout(inner, 1000.4);
+
+      await wrapped(URL_UNDER_TEST);
+
+      expect(deadlineForRequestInit(inner.mock.calls[0][1])).toBe(1000);
+    });
+
+    it("stamps nothing on an exempt request", async () => {
+      // The route must apply no deadline to MCP traffic, and it decides that
+      // from the absence of a budget in the envelope.
+      const inner = vi.fn<typeof fetch>().mockResolvedValue(new Response("{}"));
+      const wrapped = withOAuthRequestTimeout(inner, 1234, () => true);
+
+      await wrapped(URL_UNDER_TEST, { method: "POST" });
+
+      expect(deadlineForRequestInit(inner.mock.calls[0][1])).toBeUndefined();
+    });
+
+    it("reads nothing off an init that never went through the wrapper", () => {
+      expect(deadlineForRequestInit(undefined)).toBeUndefined();
+      expect(deadlineForRequestInit({})).toBeUndefined();
+      expect(deadlineForRequestInit({ method: "GET" })).toBeUndefined();
+    });
+  });
+
   describe("exemptMcpEndpoint (the transport chain's mixed traffic)", () => {
     const SERVER = "https://srv.example.com/mcp";
 

clients/web/src/test/core/mcp/remote/createRemoteFetch.test.ts

@@ -3,6 +3,7 @@ import { createRemoteFetch } from "@inspector/core/mcp/remote/createRemoteFetch.
 import {
   OAUTH_TIMEOUT_WIRE_CODE,
   OAuthRequestTimeoutError,
+  withOAuthRequestTimeout,
 } from "@inspector/core/auth/requestTimeout.js";
 
 function ok(body: object): Response {
@@ -334,4 +335,61 @@ describe("createRemoteFetch", () => {
       expect(err).not.toBeInstanceOf(OAuthRequestTimeoutError);
     });
   });
+
+  describe("carrying the caller's deadline to the route (#2319)", () => {
+    function envelopeOf(fetchFn: ReturnType<typeof vi.fn<typeof fetch>>) {
+      const init = fetchFn.mock.calls[0][1] as RequestInit;
+      return JSON.parse(init.body as string) as Record<string, unknown>;
+    }
+
+    it("puts a bounded call's budget in the envelope, not in the headers", async () => {
+      const fetchFn = vi
+        .fn<typeof fetch>()
+        .mockResolvedValue(ok(standardRemoteBody));
+      const remoteFetch = createRemoteFetch({
+        baseUrl: "http://remote.example",
+        fetchFn: (input, init) => fetchFn(input, init),
+      });
+      // Composed the way `InspectorClient` composes it: the wrapper outside,
+      // the remote fetch beneath.
+      const bounded = withOAuthRequestTimeout(remoteFetch, 7000);
+
+      await bounded("http://upstream.example/token");
+
+      const envelope = envelopeOf(fetchFn);
+      expect(envelope.timeoutMs).toBe(7000);
+      // Not a header: `headers` is re-sent verbatim to the upstream server, so
+      // a marker there would leak the Inspector's internals to a third party.
+      expect(JSON.stringify(envelope.headers)).not.toContain("7000");
+    });
+
+    it("omits the budget for an exempt call, so the route bounds nothing", async () => {
+      const fetchFn = vi
+        .fn<typeof fetch>()
+        .mockResolvedValue(ok(standardRemoteBody));
+      const remoteFetch = createRemoteFetch({
+        baseUrl: "http://remote.example",
+        fetchFn: (input, init) => fetchFn(input, init),
+      });
+      const bounded = withOAuthRequestTimeout(remoteFetch, 7000, () => true);
+
+      await bounded("http://upstream.example/mcp");
+
+      expect(envelopeOf(fetchFn)).not.toHaveProperty("timeoutMs");
+    });
+
+    it("omits the budget for a direct call that no wrapper bounded", async () => {
+      const fetchFn = vi
+        .fn<typeof fetch>()
+        .mockResolvedValue(ok(standardRemoteBody));
+      const remoteFetch = createRemoteFetch({
+        baseUrl: "http://remote.example",
+        fetchFn: (input, init) => fetchFn(input, init),
+      });
+
+      await remoteFetch("http://upstream.example/mcp");
+
+      expect(envelopeOf(fetchFn)).not.toHaveProperty("timeoutMs");
+    });
+  });
 });

clients/web/src/test/integration/mcp/remote/server-extra-coverage.test.ts

@@ -51,7 +51,6 @@ interface StartOpts {
   secretStore?: SecretStore;
   logger?: pinoType.Logger;
   seedConfig?: string;
-  fetchTimeoutMs?: number;
 }
 
 async function start(opts: StartOpts = {}): Promise<Harness> {
@@ -66,9 +65,6 @@ async function start(opts: StartOpts = {}): Promise<Harness> {
     initialConfig: { defaultEnvironment: {} },
     secretStore: opts.secretStore ?? new InMemorySecretStore(),
     logger: opts.logger,
-    ...(opts.fetchTimeoutMs !== undefined && {
-      fetchTimeoutMs: opts.fetchTimeoutMs,
-    }),
   });
   const { baseUrl, server } = await new Promise<{
     baseUrl: string;
@@ -475,19 +471,16 @@ describe("server.ts supplemental coverage", () => {
       await openStreamClosed;
     });
 
-    it("answers 504 with a typed marker when its own deadline fires (#2319)", async () => {
+    it("answers 504 with a typed marker when the request's deadline fires (#2319)", async () => {
-      // The backstop, on its own short budget so this costs milliseconds rather
+      // The caller's budget travels in the envelope, so this costs milliseconds
-      // than the production thirty seconds. It matters because this is the only
+      // rather than the production thirty seconds. The marker matters because
-      // deadline on the transport-internal discovery path, which has no
+      // a backend-enforced deadline is the only one on the transport-internal
-      // client-side wrapper — so its error has to survive the hop as something
+      // discovery path, and its error has to survive the hop as something an
-      // an `instanceof OAuthRequestTimeoutError` check can still recognize.
+      // `instanceof OAuthRequestTimeoutError` check can still recognize.
-      await stop(h);
-      h = await start({ fetchTimeoutMs: 120 });
-
       const res = await fetch(`${h.baseUrl}/api/fetch`, {
         method: "POST",
         headers: { "content-type": "application/json" },
-        body: JSON.stringify({ url: `${targetUrl}/stall` }),
+        body: JSON.stringify({ url: `${targetUrl}/stall`, timeoutMs: 120 }),
       });
 
       expect(res.status).toBe(504);
@@ -505,6 +498,65 @@ describe("server.ts supplemental coverage", () => {
       await stallClosed;
     });
 
+    it("applies no deadline to a request that carries none (#2319)", async () => {
+      // ⚠️ The regression this guards: an unconditional timer here would abort
+      // a Streamable HTTP tool call that legitimately withholds its response
+      // headers — reintroducing on the backend exactly what `exemptMcpEndpoint`
+      // prevents on the client, and reporting it as an OAuth timeout besides.
+      const caller = new AbortController();
+      const pending = fetch(`${h.baseUrl}/api/fetch`, {
+        method: "POST",
+        headers: { "content-type": "application/json" },
+        body: JSON.stringify({ url: `${targetUrl}/stall` }),
+        signal: caller.signal,
+      });
+      const settled = pending.then(
+        (r) => `responded ${r.status}`,
+        () => "aborted by us",
+      );
+
+      await vi.waitFor(() => {
+        expect(upstreamHits).toContain("/stall");
+      });
+      // Well past any budget the route might have applied on its own. Nothing
+      // should have answered: an unbounded request is still in flight.
+      await new Promise((r) => setTimeout(r, 400));
+      expect(
+        await Promise.race([settled, Promise.resolve("still pending")]),
+      ).toBe("still pending");
+
+      caller.abort();
+      expect(await settled).toBe("aborted by us");
+    });
+
+    it("ignores a non-numeric deadline rather than trusting it (#2319)", async () => {
+      const caller = new AbortController();
+      const pending = fetch(`${h.baseUrl}/api/fetch`, {
+        method: "POST",
+        headers: { "content-type": "application/json" },
+        body: JSON.stringify({
+          url: `${targetUrl}/stall`,
+          timeoutMs: "soon",
+        }),
+        signal: caller.signal,
+      });
+      const settled = pending.then(
+        (r) => `responded ${r.status}`,
+        () => "aborted by us",
+      );
+
+      await vi.waitFor(() => {
+        expect(upstreamHits).toContain("/stall");
+      });
+      await new Promise((r) => setTimeout(r, 300));
+      expect(
+        await Promise.race([settled, Promise.resolve("still pending")]),
+      ).toBe("still pending");
+
+      caller.abort();
+      expect(await settled).toBe("aborted by us");
+    });
+
     it("routes the outbound request through HTTP_PROXY (#2067)", async () => {
       // /api/fetch is the browser's ONLY way out to the network — the web
       // client's `environment.fetch` is `createRemoteFetch()`, which forwards

core/auth/index.ts

@@ -56,6 +56,7 @@ export {
   DEFAULT_OAUTH_REQUEST_TIMEOUT_MS,
   OAUTH_TIMEOUT_WIRE_CODE,
   OAuthRequestTimeoutError,
+  deadlineForRequestInit,
   exemptMcpEndpoint,
   isOAuthRequestTimeoutWire,
   withOAuthRequestTimeout,

core/auth/requestTimeout.ts

@@ -148,6 +148,39 @@ export function exemptMcpEndpoint(
   };
 }
 
+/**
+ * The deadline a wrapped call is running under, keyed by the exact `RequestInit`
+ * the wrapper handed down (#2319).
+ *
+ * This is how the budget crosses the proxy hop without travelling *upstream*.
+ * `/api/fetch` must not time every request it serves — `createRemoteFetch` also
+ * carries MCP traffic, and a Streamable HTTP tool call can legitimately withhold
+ * its response headers for minutes — so the route needs to know which requests
+ * are bounded and which are not (Copilot). A header would be the obvious
+ * carrier and is the wrong one: `serializeRequest` copies request headers
+ * verbatim into the payload and the route re-sends them to the authorization
+ * server, so a marker header would leak to a third party.
+ *
+ * A `WeakMap` keyed on the init object has neither problem. It is invisible to
+ * serialization, it needs no cast onto `RequestInit`, and it is collected with
+ * the object — the wrapper builds a fresh init per call, so there is one entry
+ * per in-flight request and nothing accumulates.
+ */
+const REQUEST_DEADLINES = new WeakMap<object, number>();
+
+/**
+ * The deadline `withOAuthRequestTimeout` stamped on this init, if any.
+ *
+ * `undefined` means the request is not bounded by this wrapper and must not be
+ * bounded by anything downstream either — an exempt MCP request reads as
+ * `undefined` because the exempt path never builds an init of its own.
+ */
+export function deadlineForRequestInit(
+  init: RequestInit | undefined,
+): number | undefined {
+  return init ? REQUEST_DEADLINES.get(init) : undefined;
+}
+
 /** The request URL, whatever form `fetch`'s first argument took. */
 function requestUrlOf(input: RequestInfo | URL): string {
   if (typeof input === "string") return input;
@@ -342,9 +375,14 @@ export function withOAuthRequestTimeout(
       }
     });
 
+    // Built once and stamped, so `createRemoteFetch` can read the budget off it
+    // and tell `/api/fetch` this particular request is bounded.
+    const nextInit: RequestInit = { ...init, signal };
+    REQUEST_DEADLINES.set(nextInit, budget);
+
     try {
       const response = await Promise.race([
-        fetchFn(input, { ...init, signal }),
+        fetchFn(input, nextInit),
         abandoned,
       ]);
       // `fetch` resolves once the response *headers* arrive, so stopping here

core/mcp/remote/createRemoteFetch.ts

@@ -22,6 +22,7 @@
  */
 
 import {
+  deadlineForRequestInit,
   isOAuthRequestTimeoutWire,
   OAuthRequestTimeoutError,
 } from "../../auth/requestTimeout.js";
@@ -154,10 +155,26 @@ export function createRemoteFetch(options: RemoteFetchOptions): typeof fetch {
           ? input.signal
           : undefined;
 
+    // #2319: tell the route whether *this* request is bounded, and by how much.
+    // The route serves MCP traffic as well as OAuth work, and a Streamable HTTP
+    // tool call can legitimately withhold its response headers for minutes — so
+    // a timer there must be per-request rather than unconditional (Copilot). An
+    // exempt request carries no deadline and the route applies none.
+    //
+    // In the JSON envelope rather than a header: `headers` is re-sent verbatim
+    // to the upstream server, so a marker header would leak to a third party.
+    const timeoutMs = deadlineForRequestInit(init);
+
     const res = await fetchFn(`${baseUrl}/api/fetch`, {
       method: "POST",
       headers: reqHeaders,
-      body: JSON.stringify({ url, method, headers, body }),
+      body: JSON.stringify({
+        url,
+        method,
+        headers,
+        body,
+        ...(timeoutMs !== undefined && { timeoutMs }),
+      }),
       signal,
     });
 

core/mcp/remote/node/server.ts

@@ -28,10 +28,7 @@ import { bodyLimit } from "hono/body-limit";
 import { watch as chokidarWatch, type FSWatcher } from "chokidar";
 import { createTransportNode } from "../../node/transport.js";
 import { createProxyFetch } from "../../node/proxyFetch.js";
-import {
+import { OAUTH_TIMEOUT_WIRE_CODE } from "../../../auth/requestTimeout.js";
-  DEFAULT_OAUTH_REQUEST_TIMEOUT_MS,
-  OAUTH_TIMEOUT_WIRE_CODE,
-} from "../../../auth/requestTimeout.js";
 import type {
   RemoteConnectRequest,
   RemoteSendRequest,
@@ -238,21 +235,6 @@ export interface RemoteServerOptions {
   /** Optional path for the user's server list file (/api/servers). Default: ~/.mcp-inspector/mcp.json */
   mcpConfigPath?: string;
 
-  /**
-   * Deadline for the outbound request `/api/fetch` makes, in milliseconds
-   * (#2319). Defaults to {@link DEFAULT_OAUTH_REQUEST_TIMEOUT_MS}, the same
-   * budget the browser-side `withOAuthRequestTimeout` uses, so the two agree on
-   * when an attempt is hopeless.
-   *
-   * This is the backstop for a client that vanishes without aborting; the
-   * primary mechanism is the caller's own signal, which `createRemoteFetch`
-   * forwards onto this hop. Exposed rather than hardcoded so a test can prove
-   * the backstop actually fires without spending thirty seconds of wall clock
-   * on it — and because a very slow authorization server is a legitimate reason
-   * to want a longer one.
-   */
-  fetchTimeoutMs?: number;
-
   /**
    * When false, the server list is read-only for this session: all
    * `/api/servers` mutations (POST/PUT/DELETE/order) are rejected with 403,
@@ -588,10 +570,6 @@ export function createRemoteApp(
   options: RemoteServerOptions,
 ): CreateRemoteAppResult {
   const dangerouslyOmitAuth = !!options.dangerouslyOmitAuth;
-  const fetchTimeoutMs = Math.max(
-    0,
-    Math.round(options.fetchTimeoutMs ?? DEFAULT_OAUTH_REQUEST_TIMEOUT_MS),
-  );
 
   // Determine auth token when auth is enabled: options > env var > generate
   const authToken = dangerouslyOmitAuth
@@ -1227,6 +1205,8 @@ export function createRemoteApp(
       method?: string;
       headers?: Record<string, string>;
       body?: string;
+      /** Per-request deadline, set only by a bounded OAuth call (#2319). */
+      timeoutMs?: number;
     };
     try {
       body = (await c.req.json()) as typeof body;
@@ -1234,7 +1214,13 @@ export function createRemoteApp(
       return c.json({ error: "Invalid JSON body" }, 400);
     }
 
-    const { url, method = "GET", headers = {}, body: reqBody } = body;
+    const {
+      url,
+      method = "GET",
+      headers = {},
+      body: reqBody,
+      timeoutMs: requestedTimeoutMs,
+    } = body;
     if (!url) {
       return c.json({ error: "Missing url" }, 400);
     }
@@ -1262,14 +1248,30 @@ export function createRemoteApp(
     // the body has been either read or explicitly cancelled — this route never
     // hands a live stream to its caller, so there is nothing left for the timer
     // to protect and nothing it could sever.
-    const timer = setTimeout(() => {
+    // ⚠️ Only the caller decides whether this request is bounded, and by how
-      deadlineFired = true;
+    // much. This route is the browser's way out to the network for MCP traffic
-      controller.abort(
+    // as well as OAuth work, so an unconditional timer here would abort a
-        new Error(
+    // Streamable HTTP tool call that legitimately withholds its response
-          `proxied request to ${url} timed out after ${fetchTimeoutMs}ms`,
+    // headers for longer than the budget — reintroducing on the backend exactly
-        ),
+    // the regression `exemptMcpEndpoint` exists to prevent on the client, and
-      );
+    // reporting it as an OAuth timeout besides (Copilot). No deadline in the
-    }, fetchTimeoutMs);
+    // envelope means no timer at all.
+    const deadlineMs =
+      typeof requestedTimeoutMs === "number" &&
+      Number.isFinite(requestedTimeoutMs)
+        ? Math.max(0, Math.round(requestedTimeoutMs))
+        : undefined;
+    const timer =
+      deadlineMs === undefined
+        ? undefined
+        : setTimeout(() => {
+            deadlineFired = true;
+            controller.abort(
+              new Error(
+                `proxied request to ${url} timed out after ${deadlineMs}ms`,
+              ),
+            );
+          }, deadlineMs);
 
     try {
       // Proxy-aware, not the bare global. This route is the browser's ONLY way
@@ -1332,14 +1334,14 @@ export function createRemoteApp(
             error: msg,
             code: OAUTH_TIMEOUT_WIRE_CODE,
             url,
-            timeoutMs: fetchTimeoutMs,
+            timeoutMs: deadlineMs,
           },
           504,
         );
       }
       return c.json({ error: msg }, 500);
     } finally {
-      clearTimeout(timer);
+      if (timer !== undefined) clearTimeout(timer);
     }
   });