Intel / post

LangChain Rebuilds MCP Support on FastMCP, Adds Stateless Spec

Source: x.com
Date: Published Sep 3, 2026

From the Daily Brief

LangChain rebuilt its Model Context Protocol integration around the newer stateless transport specification. MCP support now lives in the main LangChain package, elicitation reaches applications through interrupts, list endpoints can be cached and the adapter is built on FastMCP. Stateless transport removes the need to keep a client pinned to one server session, which makes MCP services easier to place behind load balancers and serverless infrastructure. The change is operationally useful, but it is not frictionless for existing users because package imports and adapter behavior may need updating. Teams deploying MCP at scale should review the migration details before upgrading rather than assuming this is a drop-in release.

Read the 2026-09-03 Brief →

LangChain@LangChain

ICYMI https://t.co/CDT6MZ3THa

Sep 3, 2026 View on X

Context

LangChain's post links to a write-up saying support now lives in the main library as langchain.mcp, installed with the MCP installation extra of the langchain package, instead of the separate langchain-mcp-adapters package, which the write-up describes as deprecated. LangChain says the new MCP specification dropped session-based architecture, so there is nothing left to pin to one server and a redeploy no longer kills live sessions.

Two other changes follow, according to the write-up. A tool can pause mid-run to ask the caller for information, and LangChain surfaces that as a LangGraph interrupt: the run pauses, a human answers, and execution resumes without holding an open connection. Servers can also say how long a tool catalog stays fresh, which lets a client cache tool lists instead of rediscovering tools on every run. The write-up includes a migration guide for users of the older adapters package.

Terms in this piece · Glossary

MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.

Read the source x.com