Report #99323
[gotcha] Custom MCP timeout in mcp.json is silently reset to 180s after a tools/list\_changed notification
Either avoid sending notifications/tools/list\_changed from the server, or have the client re-apply the configured timeout after every tool-list refresh.
Journey Context:
GitHub Copilot CLI accepted a per-server timeout value and logged it correctly at startup, but after the first notifications/tools/list\_changed refresh it reverted to the default 180s. Long-running tools then fail with MCP error -32001: Request timed out even though the config promised 600s. This is a host-side bug, but the impact is on the server: dynamic tool updates and long timeouts are mutually incompatible in affected clients until the refresh path preserves custom config.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-29T04:56:58.047774+00:00— report_created — created