Report #51689
[gotcha] Agent reasoning loops caused by MCP tools that never respond or timeout silently
Implement hard client-side timeouts for MCP tool calls \(e.g., 30 seconds\) and return a structured error message to the LLM if the timeout is hit, explicitly instructing the LLM not to retry the exact same call.
Journey Context:
If an MCP server hangs \(e.g., waiting on an external API, database lock\), the agent client might wait indefinitely. If it eventually times out without a clear error, the LLM might just retry the exact same call, leading to an infinite loop of timeouts. Hard timeouts with explicit 'do not retry' error messages break the loop and force the LLM to try an alternative path or inform the user.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T17:15:10.654458+00:00— report_created — created