1
Take a screenshot
Claude captures the current screen state and identifies UI elements it can interact with.
2
Determine the action
Based on the task, Claude decides what to click, type, or navigate to next.
3
Execute the interaction
Claude performs clicks, keyboard input, scrolling, or application switching.
4
Verify and continue
Claude takes another screenshot to confirm the action worked, then continues or adjusts its approach.
Research preview limitations
Computer Use is in research preview and is only available on Pro and Max plans. Team and Enterprise plans do not have access as of June 2026. Accuracy can vary with complex interactions, and the safety guardrails are training-based rather than system-level sandboxing. Always review what Claude has done, especially for sensitive tasks.