Summary
Coordinate-level Android automation, anchored to the UI the device actually reports.
Because mobile_click validates against the elements mobile_dump_ui returned, the agent taps real, interactive targets rather than guessed positions — so a flow of read-screen, tap, type, and screenshot stays grounded in the device's actual state.
What it is
A server that bridges an assistant to an Android device, exposing UI automation, app management, and screen capture so the agent can operate the phone by coordinate and by app.
What you get
- mobile_init connects to the device and must run first; mobile_dump_ui returns the screen's elements as hierarchical JSON with centre coordinates for interactive elements
- mobile_click, mobile_type, mobile_swipe, and mobile_key_press drive touch, text entry, gestures, and system buttons like BACK and HOME
- mobile_list_apps returns installed user apps and mobile_launch_app starts one by package name
- mobile_take_screenshot captures the current screen as an image for the model to view
Requirements
Runs through uvx; needs ADB installed and an Android device connected with USB debugging enabled.
Setup effort
One command — uvx android-mobile-mcp