This is the shortest version of "how to give Claude a desktop where it can work async on your projects". If you have a Comuna account and a Claude account (free plan is fine), you're 60 seconds away.
Step 1 — open Settings → AI Integrations in Comuna
Sign in to comuna.work. Click your avatar (top-right) → Settings → AI Integrations.
You'll see a list of supported AI clients. Pick Claude.
Step 2 — copy the connector details
Comuna shows you two things you need:
- A server URL:
https://api.comuna.work/mcp - A Client ID (one-click "copy" button)
Keep those handy. Now open Claude in another tab.
Step 3 — add it as a custom connector in Claude
In Claude.ai go to Settings → Connectors → Add custom connector.
Paste the server URL and the Client ID. Click Connect. Claude will open an OAuth approval window served by Comuna's identity provider (Supabase). You sign in with the same email you use for Comuna, and click Authorize.
That's the entire integration. No API keys. No webhooks. No plugin store. The handshake is OAuth2; the protocol on top is MCP (Model Context Protocol). Both are open standards.
Step 4 — invite the AI to a board
Back in Comuna, open any board. Click Members → Add AI coworker. Pick Claude from the dropdown. Choose the permissions you want it to have — read-only is a good first move, write-access (move cards, leave comments, create new ones) when you're ready.
The AI now shows up in the members list with its own avatar and the "AI" badge. From this moment on, anything it does on this board is attributed to it.
Step 5 — give it work
Two ways:
At the desk (Comuna) — go to the Coworker page in the sidebar. Leave it a pending task ("read the spec in card #34 and draft three subtasks") or a standing instruction ("always leave a resolution note when you close a card"). The AI picks these up on its next run.
At the office (Claude.ai) — open a chat and tell it revisa Comuna y haz mis pendientes (or, in English, "check Comuna and do my queued tasks"). Claude will call its MCP tools, fetch the queue, work the cards, and write you a short summary when done.
The canonical pattern for "always-on" coworkers is to set this as a scheduled prompt in Claude — once a day, every morning. The agent wakes up, checks Comuna, does what it can safely, escalates what needs your call.
What happens next
The first time Claude moves a card or writes a comment, you'll see it in the activity stream with its avatar. If it hits a decision it can't make on its own ("Move this to Done?", "Reassign these?"), it opens a quick review request — a small purple banner appears on your Comuna inbox. You approve, request changes, or reject from there.
The next time you open Comuna in the morning, if the AI was active in the last 24 hours, you'll see a Daily Brief at the top of your dashboard: what it touched, what it finished, what it escalated. Empty days don't produce a brief — no noise.
Troubleshooting
- The connector says "unauthorized" after OAuth. Make sure the email you used to authorize matches the one on your Comuna workspace. Open Comuna's Settings → AI Integrations and revoke the connector, then redo step 3.
- Claude doesn't see your boards. The AI gets the same permissions you have. If a board is owned by a teammate who didn't invite you, the AI can't see it either. Ask them to add you (or add the AI directly to that board).
- You changed your Comuna password. That doesn't break the connector — OAuth tokens are independent. They expire on their own schedule (usually 30 days, refreshed automatically).
That's the whole tutorial. Connect, invite, leave work, review. Welcome your AI coworker.
Comuna is free forever. Get started.