CursorTouch/Android-MCP

What it solves

Android-MCP provides a bridge that allows AI agents to interact directly with Android devices. It eliminates the need for complex computer-vision pipelines or pre-programmed scripts, enabling LLMs to perform real-world mobile tasks like app navigation, UI interaction, and automated QA testing using natural language.

How it works

The project operates as a Model Context Protocol (MCP) server. It integrates with Android devices via ADB (Android Debug Bridge) and the Android Accessibility API to read view hierarchies and execute actions. Because it interacts with the native UI elements, it can work with any LLM or Vision Language Model (VLM) without requiring a specialized OCR or fine-tuned CV model.

Who it’s for

This tool is designed for developers and QA engineers who want to automate Android mobile interactions using AI agents, as well as researchers exploring embodied intelligence on mobile platforms.

Highlights

• Native Integration: Uses ADB and Accessibility API for tapping, swiping, typing, and reading UI hierarchies.
• Model Agnostic: Compatible with any LLM/VLM; no custom CV or OCR pipelines required.
• Comprehensive Toolset: Includes dedicated tools for gestures, keystrokes, screen captures, device state monitoring, and shell command execution.
• Flexible Connectivity: Supports both USB and WiFi ADB connections for physical devices and emulators (Android 10+).