MCP Inspector had been shipping with no connect timeout at all; it now defaults to 30 seconds
connectionTimeout defaulted to 0, which InspectorClient.connect() read as no bound, so the only thing ending a hung handshake was the SDK's 60 s per-request timeout on initialize — which covers nothing outside that request, transport.start() included, and reports the misleading 'Request timed out'. PR #2362 (merged 2026-09-14) sets DEFAULT_CONNECTION_TIMEOUT_MS = 30000 and keys the post-timeout transport teardown to the timer actually firing rather than to connectionTimeout > 0, because with a real default the old gate would have torn down every failed connect including a recoverable 401 whose transport auth recovery holds open. Measured against a server that accepts TCP and never answers initialize: 60.5 s before, 30.4 s after.
Source
↳ Follow the thread