I’ve been working on a tool called MCPJungle - an open-source, self-hosted MCP Registry + Gateway that helps MCP clients (like Claude, Cursor, custom AI agents) connect to multiple MCP servers through a single endpoint.

MCP (Model Context Protocol) is gaining adoption as a standard for tool-calling in LLMs, but managing multiple servers (auth, tool discovery, ACLs, observability) is still a nightmare - especially across teams in an org.

MCPJungle tries to fix that:

- Expose all your MCP servers behind a single `/mcp` endpoint - Use ACLs to control which clients can view & call which MCP tools - Keep track of all your MCP clients & servers from one central place.

Individuals can run it locally for maximum privacy. Orgs deploy it as a shared gateway for all their AI agents.

It’s written in Go & distributed a single binary. You can run it via Homebrew or Docker. No auth or config needed by default - it’s meant to be friction-less for developers.

Still early, but the current version is stable and being used by a few early devs. Would love to hear feedback, critiques, or ideas. Happy to answer any questions here too.

Thanks!