Cursor's agents can now click and type across your desktop apps

On October 23 Cursor announced that "Cursor can now control your computer." The feature is called Computer Use. In Cursor's words, it "lets agents click and type across any app on your computer, not just the browser." The launch video shows an agent driving the iOS Simulator to test a feature it built.

What it does. Until now, Cursor's agents mostly wrote code, ran commands and, in some setups, used a browser. Computer Use lets an agent work the way you would: look at the screen, click, and type into a desktop app. The point Cursor is making is testing. An agent that builds an iPhone feature can open the simulator, tap through it and check that the feature works before it hands the work back.

How it starts. Cursor says the agent "uses it on its own when a task needs it." You can also ask for it directly by @-mentioning Computer Use in your prompt. To get it, Cursor says to update the app.

This isn't Cursor's first computer use. Cursor's cloud agents already run in their own virtual machines with a full desktop, where they can use a mouse and keyboard, and Cursor's docs describe a --computer-use option for self-hosted workers on macOS and Linux. On a Mac, that self-hosted version needs two macOS permissions, Accessibility and Screen Recording, before clicks and screenshots work. What's new on October 23 is the same idea inside the Cursor app on your own computer. We haven't seen Cursor say whether the desktop version uses the same helper or permissions, so don't assume it does.

Who it's for. Developers building apps with a real interface, especially mobile and desktop apps, who want the agent to prove a change works instead of just saying it does. It's also useful for anyone whose agent keeps getting stuck on a step that only a GUI can do.

The honest caveat. An agent that can click and type in any app on your machine can also click the wrong thing in any app on your machine, and Cursor says it can turn this on by itself when it decides a task needs it. When we checked on the morning of October 23 (about 9:45am PT), Cursor's changelog and docs had no page for the desktop version yet, so details like which operating systems it supports, what permissions it asks for and how to keep it to one app weren't published. Until they are, keep sensitive apps closed while an agent is working, and watch the first few runs.

MODEL has not tested Computer Use in the Cursor app yet. Everything above comes from Cursor's two launch posts on X and Cursor's existing docs, read on October 23. This is not sponsored and has no affiliate links.