MCP Mesh
A distributed AI agent framework where a decorator turns a plain function into a networked agent, and the mesh handles discovery, transport, retries and failover across languages and machines.
MCP Mesh is a distributed AI agent framework built on the idea that a specialized agent should be a small, composable dependency rather than one large model. A developer writes agent logic as plain functions decorated in Python, Java or TypeScript, and the mesh discovers, connects, heals and traces calls between them across languages, machines and clouds.
Any dependency can be a plain tool or an LLM-backed reasoning agent, and the calling code cannot tell the difference; if a reasoning agent goes down, the mesh automatically reroutes to a plain API equivalent and back again with no redeploy. It integrates with FastAPI, Spring Boot and Express, and ships a command-line tool for running, monitoring and managing an agent network, with the same code running locally, in Docker Compose, or in Kubernetes.