Your AI assistant can write code, search the web, manage files, and hold a conversation. But ask it to open an app on your phone, check a notification, or test a mobile workflow and it hits a wall.
That wall just came down.
With the Mobilerun skill on ClawHub, OpenClaw can now control real Android devices in the cloud. Not emulators. Not screenshots you paste into chat. Actual phones, running actual apps, responding to actual taps and swipes — all driven by your AI assistant.
Here's how it works, why it matters, and how to set it up in under a minute.
https://www.youtube.com/watch?v=HTUJhAVNb7k&t=4s
The stack: from prompt to pixel
You / AI Assistant
→ OpenClaw + Mobilerun Skill
→ Mobilerun Cloud API
→ Real Android Device
→ droidrun Agent Framework
→ Vision Model + Accessibility Tree
→ Screen Understanding → Action Execution
At the foundation is droidrun, an open-source mobile AI agent framework with 7.7k GitHub stars and a 91.4% score on the Android World benchmark. It's the engine that makes phone automation actually reliable.
Droidrun combines two approaches to understand what's on screen:
- Vision models — a multimodal LLM looks at the device screenshot and interprets the UI visually
- Accessibility tree — the structured representation of every UI element with properties and coordinates By fusing both signals, the agent knows what it's looking at and where to interact. It executes actions through Android's standard interfaces: tap, swipe, type, scroll, back, home.
Mobilerun wraps this into a cloud platform. No USB cable or Android SDK, no local device. Real phones in the cloud, exposed via API.
why this matters for OpenClaw users
OpenClaw is already one of the most capable AI assistant platforms. But desktop and web automation only covers half the story. Mobile is where people spend their time — and where most apps live exclusively.
The Mobilerun skill unlocks:
• QA testing — "test the signup flow on our Android app and report any issues"
• Social media management — "post this to Instagram, check engagement on my last tweet"
• Personal assistant workflows — "order me an Uber to the airport"
• Competitor monitoring — "screenshot competitor's pricing page, compare with last week"
• Cross-app automation — "copy the confirmation code from SMS and paste it into the auth screen"
Real devices. Real apps. Real state. Your assistant sees the screen, reads the UI, and takes action.
setup: 60 seconds to mobile access
- Install the skill — ClawHub → search "Mobilerun" → install. Done.
- Log in to Mobilerun Cloud — first invoke prompts authentication at mobilerun.ai
- Claim free access to Mobilerun
- Start using it:
what makes this different
| Feature | Traditional Automation | droidrun + mobilerun |
|---|---|---|
| Setup | SDK, drivers, device farm | API key |
| Scripting | xpath selectors, brittle | natural language |
| UI changes | scripts break | agent adapts |
| Cross-app | painful | native |
| Cloud devices | separate service needed | built in |
| Intelligence | none — pure execution | vision + reasoning |
get started with it
The Mobilerun skill takes 10 seconds to install and gives your OpenClaw assistant something it's never had before: hands.
Real hands, on a real phone, in the cloud.
→ clawhub.com — https://openclaw.ai/
→ mobilerun.ai — https://mobilerun.ai/
→ github.com/droidrun — https://github.com/droidrun/droidrun
Top comments (0)