mobile-mcp

Why we included this project

Agents that need to operate a phone instead of just reason about it often stall once the work moves inside a native app. This MCP server is a practical way around that: it exposes iOS and Android devices, whether emulators, simulators, or physical hardware, as tools an agent can actually drive. The agent reads the native accessibility tree for fast structured output and falls back to screenshot-based coordinate taps only when needed, which covers real work like QA automation, scripted data entry, and multi-step flows that require tapping, swiping, and filling in forms. Because it uses standard MCP, it drops into Claude Code, Codex, Gemini, and other MCP-compatible clients with minimal setup, so an existing agent stack can adopt it without reworking its orchestration.