Hey HN, author here. I've been running local LLMs with MCP tools and managing multiple servers got annoying fast. Each capability needs its own server, so at 5-10 of them you're juggling processes, wondering which one crashed, and your configs are all over the place. MCP Hangar is basically a registry that sits between your LLM client and your providers. Lazy loading so they start only when you actually call them. Health checks with circuit breaker. Works with Kubernetes (wrote an operator with CRDs) but also standalone with Docker or Podman. There's an optional auth layer if you need it. Core is Python, operator is Go, architecture is DDD with event sourcing because I wanted full audit trail for enterprise use cases. Mostly looking for feedback on edge cases I haven't thought of and what's missing for production use. Happy to answer questions.