Control any
Android phone
over HTTP
One app. 37 endpoints. Full device control.
No ADB. No root. No PC required.
v1.8.0 · AccessibilityService · CLI: cua
One app. 37 endpoints. Full device control.
No ADB. No root. No PC required.
v1.8.0 · AccessibilityService · CLI: cua
Install the app, enable the accessibility service, and start sending HTTP requests. That's it.
Download the latest release from GitHub and install on any Android device. No special permissions at install time.
Open Settings → Accessibility → Claw Use Android → toggle on. This grants the app UI tree access and input control.
The app serves a local HTTP API. Connect from any device on the same network—or tunnel through Tailscale.
curl http://phone:7333/ping
The app — service running
Settings — controlled via HTTP
Every endpoint returns JSON. Authenticate with the X-Bridge-Token header.
Returns the full UI accessibility tree as structured JSON. The foundation for AI agent perception.
curl -H "X-Bridge-Token: YOUR_TOKEN" \ http://phone:7333/screen
Captures the current screen as a JPEG image. Useful for visual AI models and debugging.
curl -H "X-Bridge-Token: YOUR_TOKEN" \ http://phone:7333/screenshot -o screen.jpg
Lists active notifications with title, text, package, and timestamp.
curl -H "X-Bridge-Token: YOUR_TOKEN" \ http://phone:7333/notifications
Check if the screen is on/off and locked/unlocked.
curl -H "X-Bridge-Token: YOUR_TOKEN" \ http://phone:7333/screen/state
Device model, OS version, display resolution, installed packages.
curl -H "X-Bridge-Token: YOUR_TOKEN" \ http://phone:7333/info
Service health: uptime, version, accessibility state, active connections.
curl -H "X-Bridge-Token: YOUR_TOKEN" \ http://phone:7333/status
Tap at screen coordinates (x, y).
curl -X POST -H "X-Bridge-Token: YOUR_TOKEN" \ -d '{"x": 540, "y": 1200}' \ http://phone:7333/tap
Click a UI element by text content, description, or resource ID. No coordinates needed.
curl -X POST -H "X-Bridge-Token: YOUR_TOKEN" \ -d '{"text": "Send"}' \ http://phone:7333/click
Type text into the currently focused input field.
curl -X POST -H "X-Bridge-Token: YOUR_TOKEN" \ -d '{"text": "Hello from Claw!"}' \ http://phone:7333/type
Swipe between two points with configurable duration.
curl -X POST -H "X-Bridge-Token: YOUR_TOKEN" \ -d '{"x1":540,"y1":1500,"x2":540,"y2":500,"duration":300}' \ http://phone:7333/swipe
Trigger global actions: back, home, recents, notifications, quick settings, power dialog.
curl -X POST -H "X-Bridge-Token: YOUR_TOKEN" \ -d '{"action": "back"}' \ http://phone:7333/global
Launch any installed app by package name.
curl -X POST -H "X-Bridge-Token: YOUR_TOKEN" \ -d '{"package": "com.android.chrome"}' \ http://phone:7333/launch
Speak text aloud using the device's text-to-speech engine.
curl -X POST -H "X-Bridge-Token: YOUR_TOKEN" \ -d '{"text": "Task complete"}' \ http://phone:7333/tts
Read clipboard contents. POST to set clipboard text.
curl -H "X-Bridge-Token: YOUR_TOKEN" \ http://phone:7333/clipboard
Capture a photo from front or rear camera. Returns JPEG.
curl -X POST -H "X-Bridge-Token: YOUR_TOKEN" \ -d '{"camera": "back"}' \ http://phone:7333/camera -o photo.jpg
Battery level, charging state, temperature, and health.
curl -H "X-Bridge-Token: YOUR_TOKEN" \ http://phone:7333/battery
Current GPS coordinates (requires location permission).
curl -H "X-Bridge-Token: YOUR_TOKEN" \ http://phone:7333/location
List device contacts with names and phone numbers.
curl -H "X-Bridge-Token: YOUR_TOKEN" \ http://phone:7333/contacts
Send an SMS message programmatically.
curl -X POST -H "X-Bridge-Token: YOUR_TOKEN" \ -d '{"to": "+1234567890", "message": "Hello!"}' \ http://phone:7333/sms
Execute a scripted sequence of actions in order. Chain taps, swipes, types, and waits into a single request.
curl -X POST -H "X-Bridge-Token: YOUR_TOKEN" \ -d '{"steps": [ {"action": "launch", "package": "com.whatsapp"}, {"action": "wait", "ms": 2000}, {"action": "click", "text": "John"}, {"action": "type", "text": "On my way!"}, {"action": "click", "text": "Send"} ]}' \ http://phone:7333/flow
Execute multiple independent commands in parallel. Returns results for each.
curl -X POST -H "X-Bridge-Token: YOUR_TOKEN" \ -d '{"commands": [ {"endpoint": "/battery"}, {"endpoint": "/wifi"}, {"endpoint": "/screen/state"} ]}' \ http://phone:7333/batch
Health check. Returns pong if the service is running.
curl http://phone:7333/ping
View or update runtime configuration (port, token, allowed origins).
curl -H "X-Bridge-Token: YOUR_TOKEN" \ http://phone:7333/config
OTA self-update. Downloads and installs the latest version without leaving the app.
curl -X POST -H "X-Bridge-Token: YOUR_TOKEN" \ http://phone:7333/install
Upload a file to the device storage.
curl -X POST -H "X-Bridge-Token: YOUR_TOKEN" \ -F "file=@local.apk" \ http://phone:7333/file/upload
▶ AI agent navigating Settings, WiFi, and browser on a Xiaomi Pad 5
Not a toy. People use this to solve actual problems with actual phones.
"81-day streak? The AI maintained it."
Your friend is learning French on Duolingo. You point Claw Use at her phone, and an AI agent reads each question, figures out the answer, taps the right options. Multiple choice, word arrangement, fill-in-the-blank — all handled. The streak lives on.
"WiFi works = full control. No screen needed."
Touch digitizer dead, USB port corroded. But WiFi connects. Install the APK via ADB once, and you've got 37 HTTP endpoints to transfer photos, read messages, back up contacts, even set up a replacement phone — all without touching the screen.
"Your phone is now a remote node on your mesh network."
Install Tailscale on the phone, enable Claw Use. Now from your laptop at a coffee shop, you can check your home security camera app, read a verification code that arrived via SMS, or trigger a smart home routine. The phone sits at home, always connected.
"Stop saying 'click the gear icon'. Just click it for them."
Mom calls: "I can't find the WiFi settings." Instead of a 20-minute phone tutorial, you SSH into her phone's Claw Use endpoint, read the screen, navigate to Settings > WiFi, connect to the right network. Done in 30 seconds. She thinks you're a wizard.
"37 endpoints. Real hardware. No emulator."
Your agent framework calls /screen to perceive, /click to act, /screenshot for vision models. It's not a simulator — it's a real Android device with a real SIM card, real apps, real notifications. The agent can send WhatsApp messages, check bank apps, take photos. Computer Use, but for phones.
"That drawer phone? It has a job now."
Old phone + charger + WiFi = a dedicated automation node. Run scheduled tasks: check app prices daily, screenshot competitor dashboards, send good-morning messages, monitor delivery status. Plug it in, forget about it. It works while you sleep.
Install the CLI tool and connect to your device immediately.
Download Latest Release# Install the CLI pip install claw-use-android # Verify connection cua ping --host phone:7333 # Read the screen cua screen --host phone:7333 # Take a screenshot cua screenshot --host phone:7333 -o screen.jpg # Tap a button cua click --text "Settings" --host phone:7333
