Claude computer use lets Claude click, type and open apps on your own computer to finish a task in Cowork or Claude Code. It's in beta on Pro and Max plans in Claude Desktop for macOS and Windows: go to Settings, then Capabilities, switch on Enable computer use, then start a Cowork session and approve each app Claude asks to use.
It can feel like magic. But it's also the Claude feature where I'd slow down the most, because Anthropic is very open that there's no sandbox between Claude and your apps. Here's how it works, how to turn it on and where I'd draw the line.
Everything below comes from Anthropic's help centre and Claude blog, read in October 2026.
Who can use Claude computer use
| Question | Answer from Anthropic |
|---|---|
| Which plans? | Pro and Max, in beta. Team and Enterprise plans don't have it at the moment |
| Which app? | Claude Desktop, inside Cowork or Claude Code |
| Which computers? | macOS and Windows |
| Does the computer need to be on? | Yes. It has to be awake, with Claude Desktop open |
| Can I keep working? | On macOS 15 or later, Claude works in background windows by default |
How to turn on computer use
- Update Claude Desktop to the latest version.
- Open the desktop app and go to Settings, then Capabilities. If there's no Computer use section there, look in Settings, then System, under This computer.
- Turn on the Enable computer use toggle.
- Open Cowork or Claude Code in the desktop app and start a session.
- Ask Claude to do something that involves an app on your computer, then approve the permission prompt for that app.
Claude asks your permission before it touches each application, and it asks again the first time in a session that a task needs your full screen.
The order Claude works in
This is the bit most guides skip, and it explains why computer use can feel slow. Anthropic says Claude picks the most precise tool first:
- Connectors. If there's a connector for the job, such as Gmail, Google Drive, Microsoft 365 or Slack, Claude uses it. That's the fastest and most reliable route.
- Browser. If there isn't, Claude may work in the browser built into Claude Desktop, or in your own Chrome through Claude in Chrome.
- Screen interaction. Only then does it use computer use to click, type and move around your desktop apps.
Anthropic gives a simple example: pulling messages through a Slack connection takes seconds, while working through Slack on screen takes much longer and goes wrong more often. So the practical tip is to connect the tools you use most and save computer use for the apps that have no connector. If you mostly work in the browser, my guide to the Claude Chrome extension may be all you need.
What it's good for
- Pulling a competitor analysis together from local files and connected tools, then formatting it as a report
- Working inside an internal dashboard or niche tool that has no connector
- Testing an app you've built in a phone simulator and noting usability problems
- Carrying on with a task while you step away, as long as the computer stays on
Pair it with Dispatch and you can hand Claude a task from your phone while it works on your desktop. I've covered that in my guide to Claude Dispatch.
Prompts to start with
Anthropic recommends simple tasks like research or organising before anything multi-step. These are the kind I'd try first:
Copy this prompt
Open the folder called Invoices 2026 on my desktop, list every PDF by date and supplier name, and put the list in a new spreadsheet. Don't move, rename or delete any files.
Copy this prompt
Open our booking dashboard in the browser, look at this week's bookings and write me a short summary of the busiest days. Read only: don't change or cancel anything.
Notice the last line of each one. Telling Claude what it must not do is the simplest safety habit you can build.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
The safety rules I'd follow
Anthropic's own guidance is unusually direct here. Claude takes screenshots of your screen to see what it's doing, which means it can see anything visible, including private documents and other people's data. Its advice is to close sensitive files and apps first and to start with apps you trust.
| Safeguard | What it does |
|---|---|
| Per-app permissions | Claude asks before using each app; investment, trading and cryptocurrency platforms are blocked by default |
| App blocklist | Any app you add is refused automatically |
| Action review | Scans for signs of prompt injection and asks before opening new apps |
| Stop button | You can stop Claude at any point |
And here's the catch people miss: an action in one app can affect another. Anthropic's example is clicking a link in your email app, which might open it in Chrome even if you never gave Claude permission to use Chrome.
Anthropic strongly advises against using computer use for managing financial accounts or investments, handling legal documents or contracts, processing medical or health information, or working in apps that hold other people's personal information. I'd add your banking app to the blocklist on day one.
Current limits
- It's a beta, and complex multi-step tasks sometimes need a second try
- Screen work is slower than connectors
- Your desktop must stay awake with Claude Desktop open
- It's Pro and Max only for now
If you'd like help deciding which jobs in your business are safe to hand to an AI agent and which aren't, that's exactly the kind of thing I help with on my work with me page.
My verdict
Claude computer use is worth switching on if you're on Pro or Max and you've got repetitive desk work in apps that don't have connectors. Connect your main tools first so Claude rarely needs the screen, start with read-only tasks, say what it mustn't touch, and keep anything financial, legal or medical well away from it. Used that way, it's a helpful extra pair of hands rather than a risk.
If you'd rather link Claude to your apps than let it drive your screen, see my guide to connecting your apps to Claude.
Frequently asked questions
Is Claude computer use free?
No. Anthropic lists computer use as a beta for Pro and Max plans. Free, Team and Enterprise plans don't have it at the moment.
How do I turn on computer use in Claude?
In Claude Desktop, go to Settings, then Capabilities (or Settings, then System, under This computer), switch on Enable computer use, then start a Cowork or Claude Code session.
Does Claude computer use work on Windows?
Yes. Anthropic says it's available in Cowork and Claude Code in Claude Desktop for both macOS and Windows.
Can I use my computer while Claude works?
On macOS 15 or later, Claude works in background windows by default, so it doesn't take over your pointer or keyboard.
Is Claude computer use safe?
Anthropic says there's no sandbox between Claude and your apps and its safeguards aren't perfect. Close sensitive apps, block banking and health apps, and start with simple tasks.
