auto-mcp

What is auto-mcp?

auto-mcp converts tools, agents, and orchestrators from CrewAI, LangGraph, Llama Index, OpenAI Agents SDK, Pydantic AI, and mcp-agent into MCP servers for standardized access via clients like Cursor and Claude Desktop.

How to use auto-mcp?

Install the package (pip install naptha-automcp), run automcp init -f <framework> in your project directory to generate run_mcp.py, edit it to import your agent and define an input schema with Pydantic, then run automcp serve -t sse or automcp serve -t stdio. The server can also be run directly with Python or uv.

Key features of auto-mcp

• Converts existing agent frameworks into MCP servers
• Supports CrewAI, LangGraph, Llama Index, OpenAI Agents SDK, Pydantic AI, and mcp-agent
• Generates run_mcp.py with STDIO and SSE transport handlers
• CLI commands for initialization and server serving
• Includes examples for each supported framework
• Can deploy to Naptha's MCPaaS platform

Use cases of auto-mcp

• Integrate existing CrewAI crews into Cursor IDE
• Expose LangGraph agents as MCP tools for Claude Desktop
• Deploy Llama Index agents via Naptha's MCPaaS platform
• Create standardized MCP endpoints for OpenAI Agents SDK orchestrators

FAQ from auto-mcp

Which agent frameworks does auto-mcp support?

CrewAI, LangGraph, Llama Index, OpenAI Agents SDK, Pydantic AI, and mcp-agent.

What transport modes are available?

STDIO (client starts the server automatically) and SSE (user starts the server separately, then client connects).

How do I avoid STDIO protocol corruption with print statements?

The generated server includes utilities to suppress warnings, and print statements in agent code can corrupt STDIO.

Where can I deploy my auto-mcp server?

You can deploy locally or use Naptha's MCPaaS platform after configuring pyproject.toml correctly.

What are the runtime dependencies?

Python, pip or uv, and the target agent framework (e.g., crewai, langgraph) along with their API keys.