Prevent MCP ClientSession hang

[njbrake]

Per https://modelcontextprotocol.io/specification/draft/basic/lifecycle#timeouts

"Implementations SHOULD establish timeouts for all sent requests, to prevent hung connections and resource exhaustion. When the request has not received a success or error response within the timeout period, the sender SHOULD issue a cancellation notification for that request and stop waiting for a response.

SDKs and other middleware SHOULD allow these timeouts to be configured on a per-request basis."

I picked 5 seconds since that's the default for SSE

[rm-openai]

@njbrake - thanks. Couple of quick questions:

1. This is different than the timeout on the SSE server? Any idea why there are two different places for timeout configuration?
2. Based on the spec - should we make this configurable? Just add client_session_timeout_seconds: int | None = 5 and use that?

[njbrake]

Thanks for the quick response @rm-openai.

@njbrake - thanks. Couple of quick questions:

1. This is different than the timeout on the SSE server? Any idea why there are two different places for timeout configuration?

Based on my understanding of the MCP python sdk, yep this is different because this ClientSession applies to both SSE and Stdio sessions. The current timeout and sse_read_timeout parameters only apply to the sse_client so they're not involved with a Stdio client, resulting in a ♾️ hang if the Stdio server dies.

The parameter that I'm adding to mcp.ClientSession is passed to https://github.com/modelcontextprotocol/python-sdk/blob/main/src/mcp/shared/session.py#L246-L263
which looks like it's the ClientSession equivalent HTTP timeout just like https://github.com/modelcontextprotocol/python-sdk/blob/main/src/mcp/client/sse.py#L25

2. Based on the spec - should we make this configurable? Just add `client_session_timeout_seconds: int | None = 5` and use that?

Sounds good to me 👍

[njbrake]

@rm-openai Ok gave a shot at implementing the timeout as you suggested, let me know what you think 🙏

[njbrake]

@rm-openai ah sorry, I see some linting etc failing, one sec I'll fix it up