android-use

Control Android devices programmatically. Tap, swipe, type, launch apps, and automate UI interactions with structured, agent-friendly output.

🤖 Agents, start here: Getting Started Guide

Why android-use?

What: A semantic layer over ADB. Simplifies and standardizes device interactions into consistent, predictable commands that agents can reliably parse and execute.

Why: Raw ADB outputs unstructured text. This tool returns structured JSON with consistent error handling, screen coordinates parsed from accessibility trees, and clear success/failure states. Screen dumps are compact and token-efficient (~50-200 tokens vs thousands for raw XML), perfect for LLM agents with context limits.

How: Perception → Action loop. Get the UI state as structured data, reason about it, execute the next action. Repeat.

Raw ADB	android-use
adb shell dumpsys window windows + parsing	android-use get-screen → structured JSON
adb shell input tap 540 960	android-use tap 540 960 with validation
Exit code 0 or manual string checking	Typed results: {success: true, data: {...}}

Install

curl -fsSL https://raw.githubusercontent.com/iurysza/android-use/main/install.sh | bash

Prerequisites: ADB installed, Android device with USB debugging enabled.

Quick Start

android-use check-device                # List devices
android-use get-screen                  # Get UI with tap coordinates
android-use tap 540 960                 # Tap at coordinates
android-use type-text "Hello"           # Type text
android-use launch-app com.android.chrome  # Launch app

Documentation

• Agent Setup Guide - Complete setup and usage guide
• Examples - Tutorials and common patterns
• Changelog

License

MIT