Can LangChain agents use MCP servers?

Yes — LangChain agents can use MCP servers via the official langchain-mcp-adapters package (released early 2025) which wraps any MCP server as a LangChain Tool. You install the adapter, point it at an MCP server's stdio command or HTTP endpoint, and the server's tools appear in the agent's toolbox alongside native LangChain tools. AgentFetch publishes a tested LangChain wrapper that takes about 10 lines to wire up. The same pattern works for LangGraph agents, CrewAI (via community adapter), and Anthropic's own mcp-agent framework. This is the main benefit of MCP — write the fetch server once and it works across Claude Desktop, Cursor, Cline, LangChain, LangGraph, mcp-agent, Zed, and Continue.dev without per-runtime integration code. For LangChain specifically, performance is comparable to native tools; the adapter adds minimal overhead because both layers are async. The one caveat: LangChain agents need to handle MCP server lifecycle (spawn, health check, kill on exit), which the adapter mostly handles but worth verifying in production deployments.