AlexGladkov/claude-in-mobile
MCP server for mobile and desktop automation — Android (via ADB), iOS Simulator (via simctl), and Desktop (Compose Multiplatform). Like Claude in Chrome but for mobile devices and desktop apps
What is mcp-devices / claude‑in‑mobile?
A modular command‑line server that lets an Anthropic Claude‑based AI agent interact with real devices – Android phones, iOS devices, browsers, desktop apps, Flutter‑Aurora, and HarmonyOS – as if they were part of its “mind”.
The project used to be shipped as a single npm package called claude-in-mobile. Starting with version 4.0 the core was renamed mcp-devices and split into a thin base plus optional platform plugins. The original all‑in‑one package still exists for users who prefer a single install.
How it works
- Install the base server (
npm i -g mcp-devices). - Add the platform(s) you need – e.g.
npm i -g @mcp-devices/plugin-androidand then enable it withmcp-devices install android. - Point your MCP‑compatible client (Claude’s “Mobile Control Protocol” client) at the server by adding a JSON entry:
{ "mcpServers": { "mobile": { "command": "mcp-devices" } } } - Ask Claude to perform actions such as “take a screenshot of my Android device”; the server forwards the request to the device via the appropriate tooling (ADB, Xcode, Chrome, etc.) and returns the result.
A native Rust CLI (mcp-devices-cli) is also provided for shell scripts, installed via Homebrew, to avoid name clashes with the npm command.
What you can do with it
| Capability | Example actions |
|---|---|
| Drive apps | Capture screenshots, simulate taps/swipes/typing, launch apps, query UI hierarchy, run shell commands, manage permissions, transfer files |
| Live debugging | Set breakpoints, step through code, inspect variables, evaluate expressions (Android JDWP, iOS LLDB) – provided by an optional debug plugin |
| Testing | Visual‑regression checks, accessibility audits, performance profiling, sensor or network condition simulation |
The core ships with ~20 built‑in tool modules (e.g., screenshot, tap, UI‑tree) and the debug plugin adds deeper inspection capabilities.
Installation options
| Edition | When to choose it | Install command |
|---|---|---|
claude-in-mobile (all‑in‑one) |
Want zero configuration, everything bundled | npm i -g claude-in-mobile |
mcp-devices (modular) |
Want a slimmer footprint and only the platforms you actually use | npm i -g mcp-devices + platform plugins |
Platform plugins are separate npm packages (@mcp-devices/plugin-android, @mcp-devices/plugin-ios, …). After installing a plugin you enable it with mcp-devices install <name>.
Who might use this
- Developers building AI‑powered assistants that need to control a phone, tablet, or desktop as part of a workflow.
- QA engineers who want to script UI interactions on real devices via an LLM.
- Researchers exploring embodied AI agents that can act on physical or simulated devices.
Where to find more info
- Docs – a set of markdown files under
docs/modules/covering each platform, the built‑in tools, and the debug plugin. - CLI help –
mcp-devices --help(Node) ormcp-devices-cli --help(Rust). - Health check –
mcp-devices doctorvalidates that required external tools (ADB, Xcode, Chrome, etc.) are present.
TL;DR
mcp-devices (formerly claude-in-mobile) is a Node‑based server that exposes a Mobile Control Protocol so Claude‑style LLM agents can drive real devices—take screenshots, tap, debug, test—through a set of optional platform plugins. Install the base, add the plugins you need, point your Claude client at it, and start issuing natural‑language commands to your hardware.
Related
- Project
- Project
- Project
- Project
- Project