fix: bump context-server request timeout from 60s to 180s

[lukemarsden]

Summary

Bumps DEFAULT_REQUEST_TIMEOUT in crates/context_server/src/client.rs from 60s to 180s. Documents the change in portingguide.md (new Critical Fix #10, plus an entry under Modified Upstream Files).

Problem

In a cold-start spec-task container, three Context server request timeout errors fire at the same instant — exactly 60s after Zed startup. All three context_servers are marked failed and their tools never appear for the session. Model reports:

<tool_use_error>Error: No such tool available: mcp__chrome-devtools__navigate_page</tool_use_error>

Why all three trip together:

• chrome-devtools and github are npx <pkg>@latest — first run does an npm download
• helixos is HTTP, but the local Helix API container is still warming up
• All three initialize requests fire in parallel during Zed boot, while CPU is contended by Zed itself, settings-sync-daemon, language servers, etc.

Reproduced in ubuntu-external-01kq5h23dvkqzqy2jmxch8hgb2. Zed.log (timestamps trimmed):

20:09:59  zed starting
20:10:59  ERROR cancelled csp request task for "initialize" id 0 which took over 60s
20:10:59  ERROR chrome-devtools context server failed to start: Context server request timeout
20:10:59  ERROR github context server failed to start: Context server request timeout
20:10:59  ERROR helixos context server failed to start: Context server request timeout

User confirmed that toggling the server off/on in Zed settings fixes it (npm has cached the package by then) — but that's a manual workaround they shouldn't need.

Fix

-const DEFAULT_REQUEST_TIMEOUT: Duration = Duration::from_secs(60);
+const DEFAULT_REQUEST_TIMEOUT: Duration = Duration::from_secs(180);

Why 180s

Cold-start npx chrome-devtools-mcp@latest on a contended container is typically 60–90s. 180s gives ~2x headroom without making genuine failures noticeably slower to surface.

Why not per-server request_timeout in settings

ContextServerSettings has no request_timeout field upstream. Adding one would touch the settings schema, the project store, and ContextServer::new_with_timeout constructor — much larger change. The constant bump is one line and benefits every MCP uniformly.

Test plan

• Build, deploy to spec-task container, start fresh task, verify mcp__chrome-devtools__list_pages works on first try (no toggling)
• Verify Zed.log no longer shows cancelled csp request task for "initialize" / Context server request timeout errors at startup
• CI green

🤖 Generated with Claude Code