ع
Learn Tracks Reference Guides Saved
feature

Computer use

Claude can open real apps, click through their screens, and check the result with its own eyes — not just edit files.

Some things you can only confirm by looking — did the button actually move, did the upload really finish, does the app open at all. Computer use lets Claude drive a real screen: open an application, click and type through its windows, and verify the outcome visually. You ask for it the same way you ask for anything in Claude Desktop — in plain language in the chat — and Claude takes over the screen, then reports what it saw. It's how Claude closes the loop on work that lives in a graphical app, not just in code and text.

Most of what Claude does is text and files — but plenty of real work only exists on a screen. Did the page render right? Did the form submit? Does the desktop app even launch? You can describe those things, but the honest way to know is to look. Computer use gives Claude that ability: it can open an application, move through its windows the way you would — clicking, typing, scrolling — and then see the result. You don’t need the terminal for any of this; you ask in plain language in the Claude Desktop chat, the same as any other request, and Claude takes the screen from there.

The point isn’t to replace editing files; it’s to close the loop on things only a graphical interface can confirm. Claude makes a change, then opens the app and checks it actually worked, rather than assuming. That’s especially valuable for flows that cross between code and a UI — a setting buried in a menu, a checkout that has to be walked end-to-end, a layout you can only judge by eye.

It’s more deliberate than the rest of Claude’s tools, and that’s by design: driving real software is powerful, so it moves carefully and you can watch every step. Use it where seeing is believing — and keep an eye on the screen when anything sensitive is on it.

why it helps Let Claude finish jobs that only a real screen can prove are done.

examples
Ask in the Desktop chat — verify in the actual app
Open the app, click through the checkout flow, and tell me if the new discount applies.
Drive a GUI step you can't script
Open the settings window and turn on dark mode, then show me a screenshot.
tips & best practices
  • Reach for it when success is something you can only see — a layout, a flow, a native app's screen.
  • It's slower and more deliberate than editing files — use it to close the loop, not for bulk work.
  • Watch what it does: clicking through real software is powerful, so keep an eye on the screen for anything sensitive.

Questions people ask

What is computer use?
Computer use lets Claude drive a real screen — open an application, click and type through its windows, and verify the result visually — instead of only editing files and text. It's how Claude finishes work that lives in a graphical app.
When should I use computer use instead of editing files?
Reach for it when success is something you can only *see* — a layout, a flow, a native app's screen. Use it to close the loop and confirm something worked, not for bulk work, since it's slower and more deliberate.
Is it safe to let Claude click through real software?
It moves carefully and deliberately by design, and you can watch every step. Because driving real software is powerful, keep an eye on the screen when anything sensitive is on it.
Can Claude take a screenshot to show me the result?
Yes. You can ask it to perform a step in an app and then show you a screenshot of the result — handy for confirming a setting changed or a flow completed.
Do I need the terminal for computer use?
No. You ask for it in plain language in the Claude Desktop chat, just like any other request, and Claude drives the screen for you. It's a Desktop-friendly capability — no commands to type.