mobile-next / mobile-mcp

MCP server for mobile automationiOS & Android

No vision model.
No image tokens.

It drives the app from the native accessibility tree.

An MCP server that lets agents and LLMs interact with native iOS and Android applications through structured accessibility snapshots, falling back to screenshots and coordinate taps only when needed. The same tools work on simulators, emulators and real devices — no XCUITest, no Espresso, no per-platform glue.

Read the source Setup guide

Two ways to read a screen

one of them is the default
Primary channel

The accessibility tree

Reads real UI elements and extracts structured data, cutting the ambiguity of screenshot-only approaches.

  • COSTNo vision model, no image tokens
  • OUTPUTStructured and deterministic
  • SCOPETaps, swipes, gestures, deep links, hardware buttons
Fallback

Screenshot + coordinates

Coordinate-based taps derived from screenshots — used only when the tree is not enough.

  • WHENElement is not exposed
  • COSTImage tokens, positional ambiguity

One API, every target

iOS Simulator

Xcode + a booted simulator

iOS Real Device

Connected over USB and trusted

Android Emulator

Android SDK + running emulator

Android Real Device

adb + USB debugging authorized

Works with Claude Code, Codex, Gemini, GitHub Copilot, Antigravity — or any MCP-compatible client.