cua-driver
- Drag gestures — The new
drag tool sends press-drag-release gestures for sliders, canvases, and other pointer targets.
- Screenshot file output —
screenshot_out_file lets the CLI write screenshots to disk, so image data does not flood the terminal as base64.
- Cursor customization — Agent cursor color and image or SVG style can be configured through the Swift API and MCP tool.
- Current reference docs — CLI and MCP reference docs are generated from source, and the text API is centered on
type_text.
- Claude Code compatibility — A Claude Code computer-use compatibility mode maps driver behavior to that client surface.
- Chrome input fixes — Chrome modifier-key leaks, silent-accept detection, and cursor animation issues were fixed.
Join our Discord