mex

Why we included this project

AI coding agents forget what they learned last session, so teams end up re-explaining how their codebase is organized and why decisions were made. mex gives those agents a persistent, repo-local memory: a deterministic code graph built with Tree-sitter indexes symbols and their relationships, while agents record what they discover into plain Markdown files. A drift checker then compares that knowledge against the source and flags claims that no longer match, so the memory stays tied to the actual implementation. The CLI also runs as an MCP server for tools like Claude Code, Codex, and Cursor, letting each session pull only the context it needs instead of scanning the whole repository. That suits teams who want agents to inherit prior discoveries rather than rediscover them from scratch.