Find your next open-source project

Search by name, purpose, or keyword — or paste a GitHub repository URL

mobile-next/mobile-mcp

Developer Tools

An MCP server that lets AI agents inspect and control iOS and Android apps on simulators, emulators and real devices.

10,647
951
TypeScript
Apache-2.0
1 year old

What it does

Mobile MCP is a Model Context Protocol server that gives an AI agent a uniform toolset for working with mobile apps. It targets iOS simulators, iOS real devices, Android emulators and Android real devices through one interface, so the agent can act without separate XCUITest, Espresso or platform-adapter code. The server exposes device listing, app install/launch/terminate, taps, swipes, text input, hardware buttons, deep links, screen recording, logs, crash reports and screenshot capture as MCP tools README.

Interaction is accessibility-first: the agent reads the native accessibility tree and can tap an element by its reference. Screenshots and coordinate-based taps are the fallback when the tree is not enough. This produces structured, repeatable output and avoids spending image tokens on every step. The intended users are developers and testers who want an agent to exercise app flows, fill forms, or explore multi-step user journeys on a connected device.

Quick start

You need Node.js 20+, Xcode command line tools for iOS work, Android Platform Tools for Android work, and an MCP-capable client such as Claude Code, Codex, Cursor, Copilot or Gemini CLI prerequisites. Start a simulator or emulator first, or connect a trusted real device.

The quick installer detects supported MCP clients on the machine and asks which ones to configure:

npx @mobilenext/mobile-mcp@latest install

Add --yes to configure all detected clients without prompting quick start. For a manual setup, add this to the client's MCP configuration:

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

Then ask the agent to "list available devices." A non-empty result confirms the server can reach your simulators, emulators and connected devices verify. Client-specific instructions for Claude Code, Codex, Cursor and others live in the repository's installation section and wiki.

Running modes and configuration

The default transport is stdio. To run a remote Streamable HTTP server, start it with --listen 3000; the MCP endpoint is then /mcp and the mode is stateless HTTP mode. Requests can be protected with a bearer token by setting MOBILEMCP_AUTH. Non-standard URL schemes are blocked by default and can be enabled with MOBILEMCP_ALLOW_UNSAFE_URLS. Anonymous telemetry via PostHog and Scarf is on by default and can be disabled with MOBILEMCP_DISABLE_TELEMETRY=1 environment variables.

For headless use, boot an Android emulator, or start the iOS Simulator with xcrun simctl boot, before the MCP server connects to it headless mode.

Constraints

iOS simulator support is listed for macOS and Linux, and Android emulators for Linux, Windows and macOS; real devices need the appropriate USB connection, trust or debugging authorization platform support. The server runs locally and communicates with the devices you connect, and its privacy policy covers data collection for the hosted cloud service privacy. The project is Apache-2.0 licensed. Device mirroring in Claude Code requires a terminal that supports the kitty graphics protocol, such as kitty or Ghostty.

HotRepo's take

The README shows a working MCP surface with real device-control tools rather than a landing page for a hosted product; the cloud device option and the companion Mobilewright test framework sit alongside the open-source server. HotRepo recorded 7,147 stars on the 2026-09-26 all-language daily snapshot and 7,645 the next day, a one-day gain of 573 stars; the project has no recorded best GitHub daily rank, so those two snapshots are the extent of the on-site signal. The jump is useful as an attention marker, not as a quality claim, and the practical question for readers is whether the accessibility-tree approach fits their app's UI structure.

Written by AI from the project's README, updated . The project's own docs are the source of truth.

You might also like

  1. jo-duchan/tapflowSelf-hosted relay and macOS agent that stream iOS simulators and Android emulators to a browser for team QA.TypeScript1,585
  2. droidrun/mobilerunAn LLM-agnostic framework that lets AI agents inspect, tap, swipe, and type on real Android and iOS devices via CLI or Python.Python10,059
  3. hridaya423/phonedriverMCP server and CLI that let AI agents read and control Android phones and iPhones with a confirmation step before payments.Rust150
  4. tester-army/e2eE2E test framework where an agent performs natural-language steps on web and mobile apps, recorded and replayed.TypeScript12,423
  5. phone-harness/mcpA hosted MCP server that lets AI agents control real iPhones and Android emulators in the cloud.Python812