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

[grll]

Describe the bug
According to the MCP specs:

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

But looking at the code the timeouts are set at the session level and not easily configured on a per-request basis as the specs recommend:

async def send_request(
   self,
   request: SendRequestT,
   result_type: type[ReceiveResultT],
) -> ReceiveResultT:
    ...
        try:
            with anyio.fail_after(
                None
                if self._read_timeout_seconds is None
                else self._read_timeout_seconds.total_seconds()
            ):
                response_or_error = await response_stream_reader.receive()
        except TimeoutError:
            raise McpError(
                ErrorData(
                    code=httpx.codes.REQUEST_TIMEOUT,
                    message=(
                        f"Timed out while waiting for response to "
                        f"{request.__class__.__name__}. Waited "
                        f"{self._read_timeout_seconds} seconds."
                    ),
                )
            )

see

Suggestion

Add an optional argument request_read_timeout_seconds and rename current _read_timeout_seconds into session_read_timeout_seconds if request_read_timeout_seconds is set it would override the session one.

I will publish a small PR in that sense.

[grll]

also maybe on another note but I find calling the argument read_timeout_seconds and using a timedelta a bit odd, I would expect an int to be passed for "seconds"

[njbrake]

Hi @grll I'm here too because we are having a similar problem in any-agent 😆 : mozilla-ai/any-agent#422

In our case, we want to have a higher timeout for the initialization, since it may involve pulling a docker image which may take longer than the timeout that we would want to use.

Maybe as a hack, we could set _session_read_timeout_seconds to be what we want when the initialization happens, and then re-set the value back to it's normal value for everything after the initialization, but it seems to defeat the purpose of there being the request_read_timeout_seconds parameter in send_request.

[grll]

@njbrake maybe not ideal but you could use a high _session_read_timeout_seconds and low request_read_timeout_seconds on every request afterward?