mcp-use is an open-source framework that provides everything you need to build, deploy, and manage AI agents powered by the Model Context Protocol (MCP). Born from the manufacturing and automation sector, it has quickly grown into a versatile platform supporting ChatGPT, Claude, and any MCP-compatible AI host. With nearly 10,000 stars and active contributions from hundreds of developers, mcp-use bridges the gap between raw MCP servers and production-grade AI applications.

• Multi-Host Support: Connect your agents to ChatGPT, Claude, or any MCP host seamlessly — the framework abstracts away host-specific quirks so you write logic once.
• Rich Ecosystem of Tools: Comes with 60+ built-in tools covering file operations, web search, code execution, API calls, and more. The mcp-use CLI scaffolds complete agent apps in seconds.
• TypeScript + Python SDKs: Full SDK support for both TypeScript and Python, with automatic TypeScript types, runtime validation, and MCP Inspector for debugging tools in real time.

The GitHub Issues are buzzing with real-world use cases, feature discussions, and bug reports from developers building production agents. Here are some highlights:

🔧 Per-Tool Authorization Layer (Issue #1257 — 5 comments)
A developer requested a per-tool authorization layer to prevent agents from accessing all tools indiscriminately. The discussion evolved into a fascinating debate about combining role-based and trust-level-based authorization:
"Great feature request — this is one of the most important missing pieces in the MCP stack... The advantage of trust levels is that you can gradually elevate permissions as agents prove themselves reliable." — @0xbrainkid

🐛 mcp-unity Server Compatibility Bug (Issue #964 — 13 comments)
A user reported that mcp-use crashes when used with the Unity MCP server, specifically when tool schemas contain $ref pointers. The root cause is that jsonschema_to_pydantic doesn't fully resolve #/properties/... refs:
"In LangChainAdapter._convert_tool, we pass inputSchema to jsonschema_to_pydantic, but that converter only resolves refs from $defs/definitions, so #/properties/... causes KeyError('position') and crashes agent initialization." — @gabrielekarra

✨ Clerk OAuth Provider Feature Request (Issue #1042 — 7 comments)
A contributor asked to add Clerk as a first-class OAuth provider alongside Auth0, WorkOS, Supabase, and Keycloak. The maintainer responded enthusiastically, asking for a working example:
"yes thank you so much, please provide also a working example of an MCP server that uses clerk oauth provider, these tests/proofs are nowadays the most important contributions you can make :)" — @pietrozullo

mcp-use stands out as one of the most complete MCP frameworks available today. Its multi-host compatibility, comprehensive tool ecosystem, and active community discussions show it's not just a wrapper around MCP — it's a full platform for building sophisticated AI agents. Whether you're prototyping a ChatGPT plugin or building a multi-agent orchestration system, mcp-use provides the building blocks with the polish you'd expect from production software.

👤 mcp-use on GitHub