Cross-platform input modes — delivery_mode applies across Linux and macOS clicks, typing, key presses, hotkeys, right-clicks, double-clicks, scrolls, and desktop actions. It controls background delivery versus foreground activation.
Honest verification — type_text, Electron targets, Chrome and Safari web content, drag and scroll, and surface-aware actions return path and verified results. Actions report verified success or explicit unverified state instead of assumed success.
Wayland, X11, and Xvnc input — GNOME and KDE Wayland window enumeration, GNOME Wayland pixel coordinates, Wayland pixel clicks, X11 foreground activation, and Xvnc keyboard input are covered.
Browser and page typing — macOS page typing over Chrome DevTools Protocol (CDP) adds insert_text and type_keystrokes; CDP target resolution is refreshed on every call.
Tool schema cleanup — Shared schema fragments align parameters across platforms, capture_mode is now ax|vision, accessibility (ax) captures omit screenshots, and structured elements include value.
More MCP clients — mcp-config can target Qwen Code, Factory Droid, and ZCode.
Join our Discord