mex
mex gives AI coding agents navigable project memory instead of one giant instruction file. `mex setup` creates a markdown scaffold: a tiny anchor file (`AGENTS.md` or `CLAUDE.md`), a `ROUTER.md` that routes tasks to the right context, a `context/` folder (architecture, stack, setup, decisions, conventions), a `patterns/` folder of reusable task guides, and an append-only decisions log written with `mex log`. The CLI then keeps it honest: eleven checkers (paths, commands, dependencies, staleness, tool-config sync, broken links and others) run with no AI tokens and report a drift score, and `mex sync` builds targeted prompts so the agent fixes only the stale parts.
Run `npx mex-agent setup` with Node.js 20 or newer (the npm package is `mex-agent` because `mex` was taken; the command is still `mex`). Setup asks which tool you use and writes the right config file: `CLAUDE.md` for Claude Code, `.cursorrules` for Cursor, `.windsurfrules`, Copilot instructions, OpenCode or `AGENTS.md` for Codex. An MCP server package exposes mex as tool calls. The README says an AST code-graph feature is an unreleased v0.7.0 preview, with the stable release at v0.6.3, and reports a community test on a homelab setup with about 60% average token reduction, which is a single independent test. It is MIT licensed.
Features
Why choose it
Trade-offs
Maintenance check
- Last commit
- 2026-10-26
- Commits, last 90 days
- 100+
- Releases, last 90 days
- 8
- Latest release
- v0.8.4 (2026-10-14)
- Open issues
- 35
- Recent issues with at least one comment
- 1/21
Oldest week on the left; the latest four weeks are orange. Source: GitHub commit activity.
Status rule: archived on GitHub = Archived; last commit within 90 days = Active; 91 to 365 days = Slowing; older = Stale. All figures come from the GitHub API on the date shown.
Most discussed open issues
- #202 MISSING_PATH: six notations and exemptions that are not drift12 comments · 2026-09-19
- #140 Field report: 40x rebuild wall-clock variance across OS users on one machine; check RSS ballooning (~12 GB); no repair path short of full re9 comments · 2026-08-25
- #231 check false positives on clean scaffolds not covered by #201-#2033 comments · 2026-09-22
- #118 Decide: do we support a second parser architecture (Prism for Ruby)?3 comments · 2026-08-04
- #93 [Go] Add code-graph extractor3 comments · 2026-07-14
Issue titles are quoted from GitHub. Open the link to check whether a problem still applies to the version you use.
Who it fits
- Your AGENTS.md or CLAUDE.md has grown too big and you want a small anchor plus routed context files.
- You want memory as plain markdown in the repo, reviewable in pull requests and diffable.
- You want drift checks that run without AI tokens and flag missing paths, stale files and broken scripts.
- You want recall across unrelated projects or conversations; mex lives inside one repository.
- You have a small project: another set of files to maintain may cost more than it saves.
- You expect the AST code-graph feature today: the README marks it as an unreleased preview.
How it differs
vs codebase-memory-mcp★ 47.7kmex is hand-curated, human-readable project knowledge such as decisions and conventions. Codebase Memory builds a code graph automatically from your source. They answer different questions and can be combined.
Compatibility
Quick start
Use cases
Alternatives
Featured in these lists
Comments
No comments yet. Be the first!