EverythingProvider timeout

EverythingProvider timeout means the MCP provider did not return a response inside the client timeout window.

Timeouts can be caused by slow upstream APIs, cold starts, network instability, or strict proxy limits.

Typical root causes

• long-running tool calls without streaming progress
• upstream dependency latency spikes
• region mismatch between client and server
• proxy timeout shorter than tool execution time
• retry storms from parallel agent calls

Immediate recovery steps

1. Retry a lightweight tool call first
2. Check status for platform health
3. Reduce payload size and test again
4. Increase timeout where your client allows it
5. Add request-level logging to measure slow segments

Server-side hardening

• set explicit upstream timeouts per dependency
• use circuit breakers for flaky APIs
• cache deterministic responses when possible
• return early with partial data for long workflows

IDE and client tuning

• avoid running many heavy tools simultaneously
• stagger retries with backoff
• run endpoint probe in Playground before publish

Verify the fix

• run three consecutive probes in Playground
• compare latency percentile, not one-off latency
• confirm no timeout errors in IDE logs

Related troubleshooting

• MCP error connection closed (-32000)
• No working IDE endpoint available
• Publishing checklist