Everything About Cursor Computer Use

On October 23, 2026, Cursor announced that its coding agents can drive desktop apps on your computer: clicking, typing, and taking screenshots. The headline example is an agent testing an app it just built in the iOS Simulator.

The agent no longer just writes the code. It checks that the app actually works.

Local agents

New: agents in the Cursor desktop app can operate desktop apps on your machine

Cloud agents

Already ran in their own virtual machines with a full desktop, and self-hosted workers can turn computer use on

• Update the Cursor desktop app
• Mention Computer Use with @ in your request, or let Cursor choose it when needed
• On macOS, approve the Cursor Computer Use helper app’s screen and accessibility permissions
• Watch the first runs closely

Exact steps can change. Check Cursor’s documentation for current setup.

• Testing iOS and Android apps in simulators
• Checking desktop apps after changes
• Clicking through sign-up and checkout flows
• Reproducing bugs from a report
• Capturing screenshots for pull requests

Build the new settings screen, run the app in the iOS Simulator, open Settings, toggle dark mode on and off, and take screenshots of both states. Report any crashes or layout problems.

• Say exactly what to test
• Say what success looks like
• Ask for screenshots as proof

• A computer-using agent can act as you in any open app
• Close apps with sensitive data before it runs
• Use a separate macOS user or test machine for heavy use
• Never let it handle passwords, banking, or payments
• Keep approvals on for actions outside your project

Treat it like letting someone else use your mouse: decide what it may touch first.

• Screen-driven testing is slower than automated tests
• Agents can misclick or misread the screen
• Complex gestures and timing can be unreliable
• It uses more of your AI usage than code-only tasks

• Keep unit and end-to-end tests for core logic
• Use computer use for visual checks and hard-to-automate flows
• Ask the agent to turn repeated manual checks into automated tests

Other coding agents, such as Claude Code and OpenAI Codex, can also test apps through browser tools and computer-use features. The idea is the same: agents that check their own work in a real interface.