Everything is keyed on the slug and task number you already use — CRD, CRD-1 — instead of internal IDs, and the workspace comes from the API key, so no call carries a workspace parameter. Beyond reading tasks, it writes back: status and description on the task, knowledge and a Mermaid architecture diagram on the project. The mock mode is worth knowing about before you wire it up — with no key you get sample data, not an error.
A server over CodeRide, a task tracker built for AI agents. Tasks and projects are addressed by readable identifiers — project slug CRD, task number CRD-1 — and the agent reads, works, and updates in place.
- The full detail of one task by its number, and the tailored prompt written for that task
- All tasks in a project, organized by status column with their order
- The next task in sequence from wherever you are
- The first task of a project plus the project details, for starting work on something new
- Every project in the workspace, with the workspace resolved from the API key rather than passed in
- Task updates that write a new description, a new status, or both
- Project-level knowledge updated in place, including an architecture diagram in Mermaid form
A CodeRide account and a workspace API key from app.coderide.ai — production keys start with CR_API_KEY_. Run it with npx -y @coderide/mcp and CODERIDE_API_KEY in the environment; the npm package is @coderide/mcp, currently 0.9.3, over stdio, and it needs Node.js and npm. Calls go to https://api.coderide.ai. Leaving the key empty or setting a placeholder starts a mock mode instead, with sample projects ABC and XYZ and tasks like ABC-1 — all nine tools respond, nothing reaches the real API.
One command plus a key — npx -y @smithery/cli install @PixdataOrg/coderide --client claude, then supply credentials