Cursor agents can now click and type in desktop apps on your own machine

Cursor's Computer Use lets its local agents operate desktop apps such as the iOS Simulator to test what they build, beyond the browser.

Cursor's agents can now click and type across desktop apps on the machine they run on, not just in the browser. In its announcement on X, Cursor says agents can "drive desktop apps like the iOS Simulator to test the features they build." If you build native or mobile apps, this targets the step where you had to tap through the Simulator yourself to check an agent's work.

AgentRiot reports that the feature is called Computer Use. The agent uses it on its own when a task needs it, or you can @-mention it in a prompt. You get it by updating Cursor. Computer use already existed for Cursor's cloud agents, which run in an isolated VM desktop, and for self-hosted workers. What's new is that it now runs on your own desktop. AlphaSignal says Cursor announced no pricing change.

That is also where the risk sits. The agent sees through screenshots, so your mail, password manager and signed-in browser share the screen with it, as a DEV Community post points out. As of AgentRiot's October 23 report, there were no release notes or desktop setup docs. Cursor hadn't said which operating systems are supported, what the desktop app asks you to approve, whether the agent asks before taking control, or which plans include it. Flavio Copes' walkthrough says that on macOS, Accessibility and Screen Recording go to a helper app named Cursor Computer Use, not to Cursor itself.

To start safely, AgentRiot suggests requesting Computer Use explicitly, watching the first runs, closing apps that hold secrets, and using a separate macOS user or machine for unattended runs. Copes suggests opening with a small inspect-only task, using a development build with disposable data.

Sources