// the find
micro/go-micro
A Go framework for services and agents
go-micro is a Go framework for building microservices and AI agents, now with a heavy pivot toward LLM-driven service generation and multi-agent orchestration. It wraps service discovery, RPC, pub/sub, and persistence behind pluggable interfaces, and adds an agent layer that turns every service endpoint into an MCP tool. Aimed at Go devs who want to avoid gluing together five separate libraries to run a distributed backend.
The MCP gateway is genuinely useful — every service endpoint automatically becomes an AI-callable tool with no extra wiring. The `plan` and `delegate` primitives are a real design decision, not just chat wrappers; agents can hand off subtasks to registered peers over RPC rather than spawning unbounded sub-agents. Hot reload with `micro run` and SSH+systemd deploy with `micro deploy` make the local-to-production story notably shorter than most Go microservice setups. Plugin architecture is real — registry, broker, store, and AI provider are all interfaces you can swap, and the repo ships production implementations for most of them.
The service-generation feature (`micro run --prompt`) is a black box — if the LLM produces broken proto or handler logic, you're debugging generated code with no audit trail, which is miserable in production incidents. mDNS as the default registry is fine for a laptop demo but falls apart in any cloud or container environment where multicast is blocked; you'll discover this at the worst time. The model layer (typed CRUD over SQLite/Postgres) is thin enough that anything beyond basic CRUD — joins, transactions, upserts — will push you off the abstraction immediately, and mixing EF-style model code with raw SQL creates two mental models in one project. The `contrib/` Python integrations (LlamaIndex, LangChain) are clearly first-cut with minimal tests; don't count on them in anything load-bearing.