Submit Log in

Mobile MCP

by Mobile Next

v1.0.0

Apache-2.0

Repository Docs npm

Featured Developer Tools

mobileiosandroidautomationtestingappium-alternative

Summary

Platform-agnostic MCP server that drives iOS and Android apps on simulators, emulators and real devices through the accessibility tree rather than screenshots.

Features

Drives apps from the native accessibility tree — no vision model, no image tokens
Same tools across iOS and Android, on simulators, emulators and real devices
Taps, swipes, gestures, hardware buttons, deep links and orientation control
App install, launch and terminate, plus screen recording
Structured, deterministic UI output for reliable data extraction
No XCUITest or Espresso knowledge required

Installation

Set up this MCP server in your favorite AI agent — copy a ready-made configuration below.

Any MCP-compatible agent

Most agents (Claude, Cursor, Windsurf, VS Code, and more) read a standard mcpServers configuration.

Open your agent's MCP configuration file.
Merge the snippet below into it.
Restart the agent — the "Mobile MCP" tools will be available.

{
  "mcpServers": {
    "mobile-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@mobilenext/mobile-mcp@latest"
      ]
    }
  }
}

Claude Code

Add the server with a single CLI command.

claude mcp add mobile-mcp -- npx -y @mobilenext/mobile-mcp@latest

Claude Desktop

Add the server to claude_desktop_config.json (Settings → Developer → Edit Config).

{
  "mcpServers": {
    "mobile-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@mobilenext/mobile-mcp@latest"
      ]
    }
  }
}

Cursor

{
  "command": "npx",
  "args": [
    "-y",
    "@mobilenext/mobile-mcp@latest"
  ]
}

VS Code

{
  "command": "npx",
  "args": [
    "-y",
    "@mobilenext/mobile-mcp@latest"
  ]
}

Description

Mobile MCP (from Mobile Next) lets an agent operate a real mobile app the way a person would — list devices, launch the app, tap, swipe, type, read the screen — without you writing XCUITest or Espresso, and without the agent needing to know which platform it is on.

Accessibility-first, not vision-first

Its defining choice is to read the native accessibility tree instead of screenshotting the device and asking a vision model what it sees. That makes runs faster and much cheaper (no image tokens), and the output is structured and deterministic: real UI elements with real identifiers, not a model's guess about a picture. Screenshots with coordinate taps remain available as a fallback for views the accessibility tree does not describe.

One API, every target

The same tools work across iOS simulators (Xcode + xcrun simctl), iOS real devices (go-ios + WebDriverAgent + tunnel), Android emulators (Android SDK + adb) and Android real devices (adb with USB debugging). Device management covers listing devices, screen size and reading or setting orientation; app management covers listing, launching and terminating apps; interaction covers taps, swipes, gestures, hardware buttons, deep links and screen recording.

What it is for

Native app test automation and data entry, scripted flows and form filling without babysitting a simulator, multi-step user journeys driven by an LLM, and general agent-based mobile interaction or data extraction. It works with Claude Code, Codex, Gemini, GitHub Copilot, Antigravity and any other MCP-compatible client.

Published as @mobilenext/mobile-mcp under the Apache-2.0 licence.