Summary of Simple MCP Client Using LangChain / TypeScript

This project demonstrates a simple Model Context Protocol (MCP) client using LangChain and TypeScript. It showcases how to integrate MCP server tools with a LangChain ReAct Agent.

The core functionality relies on the convertMcpToLangchainTools() utility from the @h1deya/langchain-mcp-tools package. This function initializes multiple MCP servers in parallel and converts their tools into an array of LangChain-compatible StructuredTool[].

The client uses Anthropic's claude-3-5-sonnet-latest as the default LLM, with commented-out code provided for OpenAI's gpt-4o. The repository also points to a more conversational MCP client and a Python equivalent for broader use cases.

To use this project, you'll need Node.js 16+, npm 7+, and optionally uv for Python-based MCP servers. An API key from Anthropic (or OpenAI) is also required. After installing dependencies and setting up the API key in the .env file, you can run the app using npm start.