# Azdroid [Health: Active]

**Category:** 🖥️ OS Automation  
**Repository:** https://github.com/az-coder-123/azdroid  
**GitHub Stars:** 6  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/azdroid

## Description
Silky-smooth, low-latency Android screen mirroring and remote control for macOS and Windows (Preview) over Wi-Fi & USB, with an MCP server for AI agents. Built with Flutter & scrcpy.

## Claude Desktop Quick Installation
No confirmed install details yet, so no config is published. Follow the setup steps in the repository above.

## Documentation

## What Azdroid MCP server does

The Azdroid MCP server gives AI assistants access to connected Android devices and emulators. Connections can use USB or Wi-Fi, while the host application runs on macOS or Windows. The project is intended for agent-driven Android testing and automation rather than only passive screen viewing.

Its workflow covers the main stages of an automated debugging task: observe the device, interact with the application, inspect failures, and verify the result. Observation includes screenshots and the Android XML semantic tree. Interaction includes taps, text entry, and gestures. Diagnostic work can use crash information and logcat traces.

The broader Azdroid desktop application also provides screen mirroring, remote control, device discovery, APK installation, file management, intent dispatching, logcat viewing, screen recording, and image cropping. These desktop features provide context for the Android workstation, while the MCP interface is described as the agent-facing layer.

## How it works

Azdroid MCP server uses standard input, or stdio, as its transport and exchanges messages through JSON-RPC 2.0. An MCP-compatible client starts the server and communicates with it locally. The server then works with Android hardware or emulators that are reachable through USB or Wi-Fi.

The README describes 62 specialized tools organized around a closed loop. An agent can first inspect the current UI and screenshot, perform an action such as tapping or entering text, collect crash or logcat evidence, and then check the resulting state. This structure suits test execution, bug triage, and tasks that require repeated observation and correction.

## Setup and configuration

The repository provides an AI Agent Setup Guide and an Autonomous Workflow Guide for configuration and operating procedures. The README identifies Claude Desktop, Cursor, and Cline as supported AI-agent clients, and describes the MCP process as stdio-based. It does not specify environment variables in the provided material.

The host platforms listed are macOS, including Apple Silicon and Intel systems running macOS 12 or newer, and Windows 10/11 x64 in preview. Android targets may be physical devices or emulators, connected over USB or Wi-Fi.

## Tools and capabilities

The documented MCP capabilities include:

- Inspecting screenshots and Android XML semantic trees
- Sending hardware taps, text input, and gestures
- Collecting crash dumps and logcat traces
- Supporting closed-loop test and verification workflows
- Working with Android devices over USB or Wi-Fi
- Exposing 62 specialized tools through stdio JSON-RPC

The desktop application additionally documents APK metadata inspection, permission flags, package launching and removal, two-way file browsing, intent dispatch, remote text injection, and hardware-backed screen recording.

## Limitations and notes

Windows support is marked as preview, while the macOS release supports Apple Silicon and Intel hardware. The supplied material does not list individual MCP tool names, detailed client configuration snippets, authentication settings, or environment variables. It also names Antigravity in the workflow description, but that client is not included in the compatible-client list here because it is not an available directory slug.

The connection model depends on the Android device or emulator being available through USB or Wi-Fi. No cloud service, analytics, or telemetry is described; the README states that video, audio, and input remain within the local network.

_Full upstream README: https://allmcps.com/mcp/azdroid/readme_