android-bridge

AI-facing interface for Android device automation via pure ADB. Designed to be called by any AI agent (Claude Code, Cursor, Hermes, etc.) through CLI or Skill.

Install

As a Skill (for AI agents)

npx skills add briqt/android-bridge

As a CLI tool (standard)

pipx install git+https://github.com/briqt/android-bridge.git

For development

git clone https://github.com/briqt/android-bridge.git
cd android-bridge
pipx install --editable .
# after changes: pipx reinstall android-bridge

Prerequisites

• Python 3.10+
• ADB (Android Debug Bridge) on PATH, or set "adb_path" in ~/.config/agent-skills/android-bridge/config.json
• Android device with USB debugging enabled (root recommended)

Usage

# Connect device (remembers for future commands)
android-bridge connect 192.168.1.100:5555

# Read screen
android-bridge snapshot                # native apps (text + interactive elements)
android-bridge dump                    # Flutter/Compose apps (full node tree with bounds)
android-bridge screenshot --out screen.png

# Interact
android-bridge tap 540 1200
android-bridge swipe 540 1500 540 500
android-bridge press back
android-bridge type "hello"
android-bridge type-cjk "中文内容" --at 638 779

# Shell
android-bridge shell "dumpsys battery"
android-bridge shell --root "cat /sys/class/power_supply/battery/capacity"

Design

• LLM is the brain, tool is the hands. No decision logic in the tool — just atomic operations (perceive, tap, swipe, type).
• Pure ADB, zero on-device dependencies. No apps, no agents, no Accessibility Service installed on the phone.
• CLI is the single source of truth. Skill is documentation; future MCP would be a protocol wrapper around CLI.