android-mcp

What it does

Android-MCP bridges AI agents to Android devices via ADB and the Accessibility API, enabling app navigation, UI interaction, and automated testing without computer vision or pretrained models. The server communicates with Android 10+ devices over USB or WiFi, providing 14 tools for tapping, swiping, text input, gesture control, screenshot capture, and shell command execution. Typical latency between actions is 2–4 seconds. It works with any language or vision model, letting you bring your own LLM to automate real-world mobile tasks.

Who it's for

QA engineers automating mobile app testing, Android developers wiring Claude into device testing workflows, and teams building AI agents that need to interact with running Android apps without programmatic UI simulation.

Common use cases

• Automated QA testing workflows that interact with app UI via taps, swipes, and form input
• Mobile app accessibility validation by navigating and reading view hierarchies
• Testing workflows that require taking screenshots and comparing visual state
• Launching apps and executing shell commands on the device

Setup pitfalls

• Device must be accessible via adb devices before the server starts; Android-MCP resolves devices lazily, but tool calls fail if no ADB device is available.
• USB debugging must be enabled on the device and connected via USB cable or WiFi ADB (port 5555) before launching the server.
• Requires Python 3.13—on Windows, Python 3.14 fails to resolve a transitive pywin32 dependency when using uvx.
• WiFi ADB requires manual adb connect host:port setup; the server accepts host/port configuration via environment variables or command-line flags.