PolyScreen
An Android-first MCP server for multi-display automation on real devices.
Works onWindowsmacOSLinuxHighlights
- Drives real Android devices over the Model Context Protocol
- Understands multiple displays, not just the primary screen
- Returns structured evidence such as screenshots and recordings
- Probes device capabilities before an agent assumes them
About
PolyScreen is an MCP server that lets an AI agent drive a real Android device. It was built for hardware that has more than one display, where “take a screenshot” is an ambiguous request until you say which screen you mean.
Install
npx polyscreen-mcpPoint any MCP client at the command and the tools appear in the client’s tool list.
Why multi-display matters
Dual-screen handhelds, foldables, and devices with an external output all behave differently per display. PolyScreen treats the display as a first-class parameter for capture, inspection, and input, so automation on the secondary screen is not guesswork.
Evidence, not vibes
Actions return structured results — screenshots, recordings, UI snapshots — so an agent’s claim about what happened can be checked instead of trusted. Capability probes report what the connected device can actually do before a run depends on it.
License
MPL-2.0, published on npm.