adb-mcp
MCP server for controlling Android devices over ADB, using direct commands and semantic accessibility selectors with a Kotlin helper APK.
README
adb-mcp
Everything was built by LLM's
MCP server for controlling an Android device over ADB. The host server is written in TypeScript and talks to a Kotlin helper APK that exposes semantic UI control through an AccessibilityService.
The project is optimized for a physical Android phone connected over USB ADB. Screenshots and coordinate gestures are intended as explicit fallbacks; normal actions use direct Android commands or semantic accessibility selectors.
Status
Prototype implementation. The TypeScript server builds and its unit tests pass. The Android helper requires a local Android SDK before it can be built.
Requirements
- Node.js 22 or newer
- npm
- Android SDK command-line tools or Android Studio
- A USB-connected Android device with ADB enabled
Install
npm install
Build and test
npm run build
npm test
To build the Android helper:
scripts/bootstrap-android-sdk.sh
npm run build:helper
If the Android SDK is already installed, set ANDROID_HOME or create helper/local.properties with:
sdk.dir=/absolute/path/to/Android/sdk
helper/local.properties is intentionally ignored by git.
Run
Build the server first:
npm run build
npm start
For an MCP client, copy mcp-config.example.json and replace /absolute/path/to/adb-mcp with this repository's absolute path.
Helper APK
Build and install the debug helper APK:
npm run build:helper
npm run install:helper
Then enable the helper AccessibilityService on the device once in Android settings.
Tools
The server exposes low-level Android control primitives such as:
- device discovery and setup status
- helper APK installation
- app launch and foreground app detection
- semantic tree lookup and node actions
- safe key events
- guarded shell execution
- bounded
run_stepscomposition
See workflows/android-device-control.md for the V1 behavior contract.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。