Connecting your repo

With a spec in hand, one command connects it to your code and your agent.

1. Open your project’s Integrations page, pick your coding tool and OS, and copy the install command. It comes pre-filled with the CLI version pinned to your project and your private token:

   npm i -g @cubedot/cli@<version>
   cubedot init --token <TOKEN>

2. Run it in your project folder.

3. Open your AI coding agent (Claude Code, say) in that same folder and reload.

Caution

The command contains your private key, so don’t share it. And copy it from the Integrations page rather than typing it: the pinned CLI version keeps your local tooling in step with your project.

init checks the connection before it writes anything, then drops three things into your repo:

• .mcp.json, pointing your agent at the CubeDot MCP server.
• Agent assets, CLAUDE.md / AGENTS.md, plus a Feature Implementation Agent and a Verifier.
• .cubedot/, the local ledger that tracks progress. If your Design System includes uploaded fonts, they land in the design cache here too.

Note

If the token is wrong or the project isn’t reachable, init tells you exactly what’s off rather than half-writing files. Using Figma? cubedot init --enable-figma also wires up the Figma MCP connection; --no-figma skips it explicitly.

To confirm it took, run cubedot status (a read-only health check), then cubedot work to see what’s ready to build. From there, the CLI reference has the rest. To disconnect a repo cleanly later, cubedot uninstall removes everything init wrote.