Fix stateless HTTP mode with elicitation and sampling

[maxisbey]

Summary

Stateless HTTP mode currently doesn't work correctly when using elicitation or sampling features.

Problem

When running in stateless HTTP mode (stateless_http=True), the SDK fails to properly handle elicitation and sampling requests. The context required for these bidirectional interactions isn't properly established in stateless mode.

Open Questions

• Stateless HTTP isn't currently part of the MCP specification
• There are upcoming spec changes around statelessness that will affect this area
• Unclear whether to fix now, wait for spec, or remove the feature entirely

Related SEPs

• SEP-1442: Make MCP Stateless (by default)
• SEP-1359: Protocol-Level Sessions for MCP

---

_{AI Disclaimer}

[jlowin]

FWIW this currently feels like a very common stumbling block. Among folks implementing MCP in Python I think this is now accepted as "just the way it works" -- if you run stateless, you give up these features -- but it's difficult to point at the cause (I'm not even sure we understood when opening jlowin/fastmcp#1156) and can create confusion as users attempt to follow best practices for deployment.

[maxisbey]

FWIW this currently feels like a very common stumbling block. Among folks implementing MCP in Python I think this is now accepted as "just the way it works" -- if you run stateless, you give up these features -- but it's difficult to point at the cause (I'm not even sure we understood when opening jlowin/fastmcp#1156) and can create confusion as users attempt to follow best practices for deployment.

(mostly just made this ticket to track a note from some meetings we have) Honestly I'm a bit confused by the stateless mode in the SDK. The Spec currently doesn't explicitly support stateless mode at all, so the implementation in the sdk is a bit of a wild west.

Maybe worth just closing this until the transports discussion has come up with the future of the spec.

Thinking as I'm writing this, I'll close this issue in favour of this one: #1696

[IsraelShok]

Hi @maxisbey
Is there an update about sampling in "stateless_http=True"?

It's a block for us to implement the sampling.

Thanks!