How to Connect Maybe Tasks to Claude via MCP

For an AI to help with your actual work instead of handing out generic advice, it needs your actual task list. In Maybe that takes one step: paste https://app.maybe.do/mcp into Claude (or Cursor, or any other MCP-capable client), sign in with your Maybe account, and the assistant can see your tasks. Access is read-only, and there are no Client IDs or secret keys to enter.

What MCP is and why your task list needs it

Model Context Protocol is an open standard that lets AI clients connect to outside data sources. Before it, discussing your workload with an assistant meant pasting the list into the chat by hand — and a day later that paste was stale. MCP removes the copy-paste: the assistant pulls your current tasks mid-conversation.

Maybe runs its own MCP server, so the integration isn't tied to one app. Claude, Cursor and any other client that speaks the protocol connect to the same address.

The assistant gets four tools: tasks by status (active, later, completed, archived), the contents of your current In Progress session, search across task titles, and stats for a date range. That's enough to clear your plan before a pass, break down a stuck anchor, and review your week.

How to connect Maybe to Claude

The address is the same everywhere: https://app.maybe.do/mcp. You can copy it straight from the app — Settings → AI (MCP) → "Connect an assistant", where there's a copy button.

Then, in Claude:

1. Open Settings → Connectors → Add custom connector.
2. Paste https://app.maybe.do/mcp and tap Add.
3. In the connectors list, tap Connect.
4. Sign in with your Maybe account (Apple or Google) and confirm access.

That's it — the task tools show up in your chat. Claude calls them on its own whenever a question needs them, so there are no special commands to memorise.

This screen shows everything the assistant gets: four tools, all of them read-only. Nothing resembling "create" or "delete" is in the list.

Other clients: Cursor and the rest

Same routine: find wherever the app asks you to add an MCP server, paste the same address, and sign in. There's no client-specific configuration — the server announces its own capabilities on connect.

What to ask the assistant if you work by FVP

Maybe is built around the FVP method: you move down your plan from the anchor and pick a task only if you can say "I want to do this more than the one I already picked." No AI can answer that for you — the wanting is yours, and that's the whole point of the method. What it does help with is everything around the pass: preparing the list, cracking the anchor, and seeing what came of it.

Clear the plan before a pass. "Group my active tasks by topic and show me where several entries mean the same thing." The shorter and more honest the list, the faster you move through it on "want." A plan that's grown to a hundred lines turns into a task graveyard, and the pass gets exhausting before you ever reach the anchor.

Deal with the anchor. The first task in your plan is the oldest one, the one you can no longer ignore. It usually sits there because it's too big. "Here's my first active task — break it into 3–4 small steps I can actually start with." A small step is easier to want.

See what's stalling. "Which tasks have been sitting in my active list the longest, and what do they have in common?" Tasks you skip over on every pass are a signal: either the wording is vague, or they belong in Later.

Get the current chain moving. "What's in progress right now? Sketch a 20-minute plan for the first one." You built the order in In Progress yourself — the assistant doesn't reshuffle it, it just helps you start the next thing faster.

Review your week. "How many tasks did I close in the last 7 days and what were they? Which themes repeat?" What you finished shows what you actually had energy for — more interesting than a count of closed items.

Pull context for a meeting. "Find all my tasks about project X and turn them into a short summary." Search runs over your own titles, so the summary comes from real work rather than generalities.

Prompts you can paste as-is:

• "Group my active tasks by topic and find duplicates."
• "Take the first task in my plan and break it into small steps."
• "What's in progress right now? What's the easiest thing to start with?"
• "What did I complete this week? Break it down by day and by theme."
• "Which active tasks should have moved to Later long ago?"

How safe is this

That's the first question when connecting an AI to personal data, and the answer is short: the assistant can only read.

• Read-only. No write tools exist — the AI physically cannot create, change or delete a task.
• Your data only. Every request is filtered by the user id inside the token. Other people's data, passwords and tokens are out of reach by design.
• Signed in with your account. Authorisation runs over OAuth 2.1 with PKCE, through the same Apple/Google sign-in as the app. The address you paste is public — no secrets to hand over.
• Revoked in one click. Remove the connector in your AI client's settings and access stops.

Wrapping up

One address field, a sign-in you already have — and the assistant stops giving generic advice, because it sees your real plan. The "want" call stays yours: the AI clears the list before a pass, helps crack the anchor, and shows what came out of the week. Your tasks stay exactly where they were — in Maybe.