@archme/cli

Command-line companion for ArchMe — maps your codebase into a living architecture diagram and keeps it current as you commit.

npm install -g @archme/cli

Requires Node 20+ and Claude Code installed and signed in — scans and updates run on your Claude subscription, on your machine. Your code is never sent to ArchMe's servers; only the resulting architecture is.

1. Scan your repo

cd into the repository you want mapped, then:

archme init

This signs you in, creates a project named after the repo, writes the agent config (.mcp.json, a CLAUDE.md block, the architecture skill), tells you what the scan will cost before spending it, and runs the scan. It prints a link to your board — components appear on it as they're written.

2. Keep it up to date

archme init also starts watching the repo, so every push queues an architecture update — but nothing processes that queue until you run the agent:

archme daemon start

Installs the daemon as a background service (launchd/systemd), enables it at login, and starts it. Run it on a machine that stays on. Each update counts tokens against your Claude account, the same way the scan did.

Watching a repo needs the ArchMe GitHub App installed on it; archme init tells you if that step is still missing. Not up for hosting the agent yourself? Get in touch at archme.io and we'll run it for you.

3. Remove it

archme uninstall

From inside the repo. It lists everything it will change and asks before doing any of it:

• The agent config — the archme MCP server from .mcp.json, the ArchMe block from CLAUDE.md, the architecture skill. Your own config in those files is left alone, and a file we created is deleted rather than left empty.
• Watching — so commits stop updating the architecture. Removing the config on its own would leave the board moving while looking uninstalled, which is why this is the default; --keep-watching opts out.
• The daemon — but only if no other repo is left for it to serve. It's one service per account, so uninstalling one repo never stops the others.

Your architecture on archme.io and this machine's login survive (archme logout for the latter). The ArchMe GitHub App installation is org-wide, so it stays — remove it on github.com if you want it gone.

Other commands

Every command takes --help. Useful flags:

• archme init --skip-scan — wire up the repo without scanning; --force re-scans a project that already has an architecture; --yes runs unattended (scans are interactive by default, so you can steer them as they go).
• archme daemon status — is it installed, running, and signed in? archme daemon logs -f follows the log.
• archme uninstall --keep-watching — removes the agent config but leaves the repo watched and the daemon running (see above); --yes skips the confirmation. Run it from anywhere in the repo — it resolves the repo root.
• archme analyze outline <file> — JSON symbol outline for one file; repo outlines everything, diff <range> only what a git range touched. Supports TypeScript, TSX, JavaScript, Python, Go, Java, Ruby, Rust, Swift, and Kotlin. Grammars download on first use to ~/.archme/tools/grammars/.

Notes

• Credentials: stored in the macOS Keychain when available, otherwise ~/.archme/credentials.json (mode 0600). keytar is an optional dependency — if its native build fails, everything still works via the file store.
• Configuration: no env setup needed against archme.io; the CLI never reads .env files. Self-hosted deployments can override SUPABASE_URL, SUPABASE_ANON_KEY, ARCHME_DAEMON_CLIENT_ID, ARCHME_MCP_BASE_URL, and CLONE_BASE_DIR (daemon only, defaults to ~/.archme-job-runner/clones).
• Agent methodology — how an agent scans a repo, anchors components to code, and writes architecture updates — is in the SKILL.md bundled with this package. archme init installs it into your repo for you.

License

Proprietary — see the bundled LICENSE. Use is granted against the archme.io service; bundled open-source components keep their own terms.