Claw Use Android

What This Skill Does

Claw Use Android (CUA) is a powerful bridge between AI agents and real Android hardware. By running as a background service on your phone, it bypasses the need for ADB, rooting, or constant PC connectivity. The skill exposes a set of HTTP endpoints that allow agents to observe the screen through a semantic UI tree, take high-quality snapshots, and execute precise actions—ranging from simple taps and text input to complex multi-step swipes and application launches. It transforms your mobile device into an interactive laboratory for your AI agent, enabling it to 'see' what you see and 'interact' as effectively as a human user.

Installation

To begin, ensure you have the cua CLI installed on your local environment. Follow these steps:

1. Install the Claw Use Android APK directly onto your Android device.
2. Open the app and enable the required Accessibility Service in your phone settings to allow the agent to read screen content and trigger UI interactions.
3. Register your device from your local terminal using: cua add <device_name> <ip_address> <token>.
4. Verify connectivity by running cua ping. Once successful, you are ready to integrate your agent with the device via the Unified API.

Use Cases

This skill is ideal for automating repetitive mobile tasks, managing app workflows, or conducting mobile quality assurance. You can use it to automate food ordering, manage notifications across different apps, perform regression testing by comparing UI states, or transcribe and respond to messages in social apps. The new Flow-First mechanism allows you to define reusable patterns in flows.md, enabling the agent to execute complex sequences like 'log in to the app and navigate to settings' in milliseconds without redundant LLM reasoning.

Example Prompts

1. "Open Duolingo and check if I have completed my daily streak today. If not, click on the current lesson and start it."
2. "Look at the notification shade. If there is a message from Slack, read it and summarize the latest update for me."
3. "Go to the Settings app, find the display brightness, and turn it up to 80%."

Tips & Limitations

To maximize the efficiency of your agent, always leverage the semantic UI tree provided by /screen -c. This returns a compact view of interactive elements, reducing token consumption and noise. Remember the Flow-First Principle: whenever you find yourself repeating the same screen to act sequence, define it as a flow in your flows.md file. This shifts processing from the LLM to the device-side execution, resulting in sub-second response times. Be aware that the reliability of this skill depends heavily on the Android Accessibility Service; if the UI structure is overly complex or uses non-standard components, some elements may require coordinates (tap) rather than semantic ref IDs. Always have a fallback mechanism for when elements are not found.