Myco
Project & team intelligence for your agents
Myco is an open-source project- and team-intelligence layer for AI coding agents. It captures the decisions, gotchas, plans, and conventions that live in a team's head — and exposes them to every agent (Claude Code, Codex, Cursor, opencode, and others) through MCP, so agents stay consistent with how the team actually works.
What it does
Myco runs locally as a vault on the developer's machine and exposes a set of MCP tools — search, recall, save, plans, sessions, and a "canopy map" of the codebase. Agents call these tools to orient in unfamiliar code, retrieve prior decisions, persist new spores (durable lessons), and avoid relearning the same gotchas every session.
Why it exists
Without persistent context, every agent session starts cold. Teams either copy-paste long preambles into each prompt, watch agents repeatedly rediscover the same constraints, or write rules files that decay. Myco closes that loop: lessons captured by one agent in one session are retrievable by every agent in every future session.
Stack
TypeScript and Node, MCP server with the official Claude Agent SDK transport, sqlite-vec for embeddings, plus a Canopy indexer that summarizes files for fast orientation.
- MCP
- Claude Agent SDK
- TypeScript
- Node
- sqlite-vec
Ready to put agents to work?
Whether it's an AI integration, a platform rebuild, or a new secure network, we show up and ship.
[email protected]