What it does
MCPAdapt bridges Model Context Protocol servers into framework-specific tools for agentic workflows. It adapts tools from 650+ MCP servers for use within langchain, crewAI, google-genai, and smolagents without requiring each framework to implement MCP natively. Developers specify an MCP server (via stdio command or Server-Sent Events), instantiate MCPAdapt with their chosen adapter, and receive a tool collection ready for agent use. Single or multiple servers can be chained together.
Who it's for
ML engineers and backend developers building agents who want to leverage MCP servers across their chosen framework. Useful for teams already using langchain or crewAI who discover relevant MCP servers and need a lightweight bridge without rewriting agent scaffolding.
Common use cases
- Integrate specialized MCP servers into existing langchain or crewAI agents
- Combine multiple MCP servers in a single agent workflow
- Access curated MCP catalogs (glama.ai, smithery.ai) from your preferred framework
- Async agents that benefit from concurrent MCP tool execution
Setup pitfalls
- Install framework-specific extras:
pip install mcpadapt[langchain], not the bare package. - Security warning: only connect to trusted MCP servers, especially over SSE (Server-Sent Events)—untrusted servers gain tool execution scope.
- Requires proper MCP server configuration (
StdioServerParameterswith command and args, or SSE URL). - Reads and writes the filesystem—needs appropriate sandbox or permission controls in constrained environments.