Discussion: Multi-agent discovery for codebase context systems? #33
Description
Hey @PatrickSys 👋
Really interesting project — the idea of a "second brain" context layer for AI agents resonates a lot with what we're seeing in the multi-agent space.
I'm working on AGENIUM, an open-source naming & discovery layer for AI agents and MCP servers. Think DNS for agents — named identities (agent://your-agent), capability manifests, mTLS auth, runtime discovery.
One pattern we keep running into: when you have multiple specialized agents (or MCP servers), they need a way to find and verify each other at runtime. For a codebase-context system, imagine an agent that can dynamically discover which analysis tools are available — not through hardcoded config, but through protocol-level discovery.
Quick question: In your multi-agent scenarios, how do agents currently discover what other tools/contexts are available? Is it all manually wired, or have you explored any dynamic discovery patterns?
We're running a short research project — interviewing 5 MCP ecosystem developers about discovery pain points. Would you be open to a 15-min chat (or even async text)? No sales pitch — genuinely mapping the problem space.
Related discussion on MCP spec repo: modelcontextprotocol/modelcontextprotocol#2257
Thanks for building useful stuff! 🚀