mobile-mcp
Mobile MCP is a Model Context Protocol server for mobile automation and development through one platform-agnostic interface, so agents do not need separate iOS and Android code paths. It lets an agent interact with native apps and devices either through structured accessibility snapshots or through coordinate-based taps based on screenshots. The README's use cases include native app automation for testing or data entry, scripted flows and form interactions without manually driving simulators, emulators or real devices, and multi-step user journeys driven by an LLM.
Its tools fall into groups: device management (list devices, screen size, orientation), app management (list, launch, terminate, install from .apk, .ipa, .app or .zip, uninstall), screen interaction (screenshots, listing UI elements with coordinates, tap, double-tap, long press, swipe) and input and navigation (type text, press hardware buttons, open URLs). On iOS it supports simulators and real devices through native accessibility and WebDriverAgent; on Android it supports emulators and real devices through ADB and UI Automator. It prefers accessibility trees, which need no computer-vision model, and falls back to screenshot analysis when accessibility labels are missing. The project is Apache-2.0 licensed with a public roadmap and wiki.