Can AI Manage Your Things 3 Tasks?

Short answer: Partly, and only on a Mac you’re sitting at. An assistant can push new items into Things 3 through Shortcuts, the things:/// URL scheme, or an AppleScript bridge. Reading your list, reasoning about it, and acting on it from anywhere is the part that doesn’t work, because Things has no network API for an assistant to reach. An assistant that triages your inbox needs a task manager built to be read and written by one.

What “AI managing your tasks” requires

Four different things get asked for under one name, and they get harder in order. Capture is write-only and one-way: “Add ‘renew passport’ for Thursday.” Reading comes next, and to answer “What’s on my Today list?” the assistant has to see your data.

Triage is read and write with judgement in between: “Sort my inbox into projects and schedule the urgent ones.” Standing automation is all of that with nobody at the keyboard: “Every Friday, move anything I didn’t finish into next week.”

Things 3 does capture. It struggles past it.

What works with Things 3 today

Capture is solid. Things registers things:///add?title=... and ships Shortcuts actions on Apple platforms, so any assistant that can trigger a Shortcut or open a URL on your device can get an item into your Inbox.

Things also gives you a private email address that drops messages into your Inbox. An assistant that can send email captures into Things from anywhere, the one route that doesn’t need your Mac.

On macOS, AppleScript can read and write the Things database, so a local MCP server or script gives a desktop assistant real access. That’s the closest thing to full control, and does Things 3 have an MCP server? goes into it properly.

What doesn’t

claude.ai in a browser, ChatGPT, or any agent running on a server has no route to your Things data. No API to call, no hosted server to connect to. Everything above depends on code running on your own awake, unlocked Mac.

Those automation surfaces are all Apple technologies too. If you’ve already hit the Windows wall or the missing web app, the AI story ends in the same place.

A weekly review your assistant runs on its own needs something reachable when you aren’t there. A laptop that might be shut is not that. And tools that read Things.sqlite3 directly are reading an undocumented schema Cultured Code is free to change.

None of this is a criticism of Things. It’s a beautifully made single-player app for Apple devices, built before assistants needed programmatic access to anything, and it has never claimed otherwise.

What to look for instead

If reading, triage, or standing automation is what you want, the requirement is specific: a task manager with a hosted MCP server and a documented model. Not an “AI features” badge. Most of those are a text box that suggests subtasks. The test is whether an assistant you didn’t install can act on your real list.

Two things follow from that. It has to be hosted rather than local, so the assistant reaches your tasks without a particular machine being awake.

And it needs real views rather than just fields. An assistant triaging into Inbox / Today / Upcoming / Anytime / Someday is making a meaningful decision, while one triaging into an untyped list of labels is guessing.

How Done Bear does it

Done Bear hosts an MCP server at https://mcp.donebear.com/mcp. Connect it once with OAuth in Claude, ChatGPT, or any MCP client, and the assistant can read your list, capture, triage your inbox into the GTD views, plan a day around what’s there, and mark things done. Writes need your approval, and access is revocable.

For the standing-automation case there’s a CLI and a GraphQL API, so a cron job or an agent can run your Friday sweep without you present. Can Claude manage my to-do list? walks through what that looks like in practice.

If capture is all you want, stay on Things 3. Shortcuts and Mail to Things already do it well, and switching apps to gain it would be a bad trade. The case for moving starts at reading and triage.

Bringing your tasks over

donebear import things

Run it on the Mac where Things is installed, and areas, projects, and tasks come across with their dates. The import guide has the details, and Done Bear is free to start so you can run both for a week first.