Best for
- Teams building React Native, Expo, Flutter, or native apps
- Teams validating apps on local devices, simulators, emulators, or CI
- QA and development workflows where an agent chooses actions at runtime
I
io.github.callstackincubator/agent-device
Community
Let AI agents inspect, control, and verify apps.
GitHub source repository ↗
59FMRS · C
An open-source MIT-licensed device automation CLI from Callstack for agent-driven app inspection, control, verification, debugging, and evidence capture. The manifest lists version 0.20.0, an npm package, and stdio transport; the source does not provide specific MCP tool identifiers or a client configuration example.
agent-device is a command-line tool that lets AI agents inspect, control, debug, and verify apps on iOS, Android, tvOS, Android TV, Amazon Vega OS TV, the web, macOS, and Linux. It supports accessibility snapshots, element targeting, device actions, screenshots, video, logs, traces, network data, performance samples, crash details, and React profiles where supported.
Tools
The tool list has not been reviewed yet.
Setup
Node.js 22.12 or newer is required; web automation requires Node.js 24 or newer. Install the CLI with npm install -g agent-device@latest, then run agent-device doctor to check setup. The server manifest declares npm package agent-device version 0.20.0 with stdio transport.
Fit and risk
Not for
- Teams seeking a complete replacement for conventional maintained test suites
- Workflows requiring physical Fire TV, capture, or Vega selector backends
- Users without an available supported device, simulator, emulator, or desktop target
Required permissions
- Access to and control of connected iOS, Android, TV, desktop, or other supported device environments
- Reading accessibility trees, logs, network data, performance information, and crash details where supported
- Saving screenshots, videos, traces, and other verification evidence when requested
Risks and side effects
- Device actions can change application state or perform test operations
- Element targeting and verification may be less reliable when accessibility data is poor
- Earlier @eN references may no longer identify the same element; use references from the latest output
- Vega OS support is currently VVD-only; physical Fire TV, capture, and selector backends are unsupported
Troubleshooting
- Run agent-device doctor first to check setup
- Verify that the Node.js version meets the requirements for the target
- Run agent-device help workflow for current CLI behavior and workflow guides
- Check that the target device, simulator, or emulator is running and accessible
- Prefer references from the latest snapshot or settle diff; use screenshots or video when accessibility data is insufficient
Use cases
Let coding agents operate running apps and verify code changes
Run mobile end-to-end tests and repeatable workflows
Diagnose issues with screenshots, logs, traces, and performance data
Save .ad scripts or export strict Maestro YAML for replay
Supported clients
Supported clients have not been confirmed yet.