LangChain has moved Model Context Protocol support into its main Python package and updated it for MCP’s new stateless design. The change removes the need to keep each client tied to the server instance that created its session, simplifying deployment and recovery when servers restart.
The integration uses FastMCP to negotiate the protocol version for each connection, so older session-based servers can continue working alongside new stateless ones. It also supports cached tool lists, avoiding a discovery request before every agent run when a server says its catalog is still fresh.
A new elicitation mechanism lets a tool pause to ask for missing information or approval. LangChain exposes that pause as a LangGraph interrupt, allowing a reviewer to confirm a destructive action or supply a parameter before the same request resumes. Because the interaction is stateless, the server does not need to hold a connection open while it waits.
The feature requires `langchain[mcp]>=1.4.0` and remains in beta, so its API may change. Python support is available now, with TypeScript planned. For developers, the main benefit is a more scalable tool connection that still provides a clear checkpoint for human decisions.