The shortest version
- In Comuna: Settings → AI Integrations → Claude → copy the server URL and Client ID.
- In Claude.ai: Settings → Connectors → Add custom connector → paste those values → Connect → Authorize.
- In any Comuna board: Members → Add AI coworker → Claude → choose permissions.
- In Claude.ai: tell it "check Comuna and do my queued tasks".
That's it. No API keys, no webhooks, no plugins, no proxies. The whole handshake is OAuth2; the protocol on top is MCP (Model Context Protocol), an open standard built by Anthropic.
A detailed walkthrough with screenshots lives in Connect Claude to your project board in 60 seconds.
What Claude can do once connected
Through about 80 MCP tools, Claude can:
- Read your boards, columns, labels, members
- Read any card's full content (description, dates, checklist, comments, history)
- Create, move, complete and delete cards
- Manage labels, assignees, dates, dependencies, checklists, attachments
- Read and write notes, wiki pages, goals, key results, milestones
- Send chat messages and DMs in project channels
- Propose new cards for you to accept (light-touch escalation)
- Surface attention items across all your boards (stale cards, overdue dates, unanswered reviews)
- Report its own progress so you see "Claude · working on Card #34"
All actions are attributed to Claude in the activity feed. You can revoke the connector at any time from Comuna or Claude.
Permissions
The AI gets the same permissions you have, scoped per board. On a board with read-only members, the AI can read but not write. On a board where you have admin rights, the AI can do everything you can — unless you explicitly limit it via the AI Coworker permissions panel.
When the AI is about to do something that needs your judgment ("delete this card?", "reassign these three?"), it doesn't. It opens a review request and waits.
Cost
Zero on Comuna's side. Connecting and using Claude is free in Comuna forever.
The cost on Anthropic's side is whatever your Claude plan covers. The free tier of Claude works for light usage; Claude Pro ($20/month at the time of writing) gives you considerably more headroom. Comuna does not proxy tokens, does not take a cut, does not add charges per AI call.
Privacy
When Claude reads a card on your behalf, that card is sent to Anthropic via MCP for that specific call. Anthropic handles it according to their own data policy — they do not train on consumer-tier API calls by default. We don't proxy the data through ourselves, we don't keep a copy, we don't replay it.
The connector OAuth tokens are stored encrypted in Supabase, scoped to your user. Revoke them any time in Comuna's Settings → AI Integrations or in Claude's Connectors panel.
Troubleshooting
- Connector says "unauthorized" after OAuth. Make sure the email you used to authorize matches the one on your Comuna workspace. Revoke and redo step 2.
- Claude doesn't see a board. The AI inherits your permissions. If the board owner didn't invite you, the AI can't see it either. Ask to be added.
- You changed your Comuna password. OAuth tokens are independent; they don't break. They expire on their own schedule (~30 days, auto-refreshed).
- You want to disconnect. Comuna → Settings → AI Integrations → Disconnect. Or in Claude.ai → Settings → Connectors → Remove. Either side cuts the link.
Other AI clients
ChatGPT works the same way (Settings → Apps → Developer Mode → Create — requires Plus / Pro plan). Gemini and Cursor are on the roadmap; the MCP server already supports them — we're waiting for those clients to GA their MCP connector UIs.
Spin up a workspace and connect Claude.