mcp-use

The fast, Pythonic way to build MCP servers and clients.

An open-source Python framework for building Model Context Protocol servers and clients. You declare a tool as a Python function and FastMCP generates the schema, validation, and protocol plumbing, and handles transport, auth, and the connection lifecycle for you. FastMCP 1.0 was folded into the official MCP Python SDK; v2 is the actively developed standalone superset. The framework is free and Apache-2.0; Prefect's Horizon platform adds managed hosting and enterprise governance.

1000+ authenticated toolkits for AI agents, via MCP or API.

Tool layer that gives AI agents secure, managed access to 1000+ toolkits and 20,000+ tools across SaaS apps, exposed via MCP or direct APIs. It handles authentication, tool search, and context management, and its MCP Gateway hands each team a managed endpoint to paste into Claude, Cursor, or ChatGPT. The core is MIT-licensed; a free tier covers 20K tool calls per month.

Build and deploy AI agents as APIs, with MCP tools built in.

Toolhouse is a backend-as-a-service for AI agents: define an agent in plain language or code and deploy it as an API in one command. Agents connect automatically to the Toolhouse MCP server, which provides RAG, memory, code execution, browser use, and 1,000+ tool integrations over the Model Context Protocol.

The MCP runtime for secure, authenticated agent tool-calling.

A platform that lets AI agents take real, authenticated actions across business systems like Gmail, GitHub, Slack, and Notion. Arcade handles the hard part of agent tool-use — secure, per-user authorization — so a tool call runs with the actual end user's permissions rather than a shared API key. It ships agent-optimized prebuilt tools plus an open framework for building custom ones, exposed over MCP. A free tier is available for getting started.