native-devtools
MCP server for native app testing — screenshot, OCR, click, type, find_text, template matching. macOS, Windows & Android. Works with Claude, Cursor, and any MCP client.
claude mcp add --transport stdio sh3ll3x3c-native-devtools-mcp npx -y native-devtools-mcp
How to use
native-devtools-mcp is an MCP server that provides Computer Use capabilities for native desktop apps and Android devices, including screenshot capture, OCR, input simulation, and window management. It exposes core actions like take_screenshot, click, type_text, find_text, element_at_point, and image-based template matching (load_image/find_image), enabling AI agents to observe and interact with real UI across platforms. The server is designed to be run locally (no data leaves your machine by default) and is compatible with MCP clients such as Claude Desktop, Claude Code, and Cursor. To get started, run the server via npx or install globally, then use the included setup workflow or manual configuration to register the MCP server with your AI client. The AGENTS.md resource is provided for agent-specific prompts and intent definitions to maximize compatibility with LLMs.
How to install
Prerequisites:
- Node.js 18+ and npm (for npx usage or global installation)
- Optional: Rust toolchain if building from source (cargo) is desired
Installation options:
Option 1 — Run with npx (no install required):
npx -y native-devtools-mcp
Option 2 — Global install (preferred for repeated use):
npm install -g native-devtools-mcp
Option 3 — Build from source (Rust):
<details> <summary>Click to expand build steps</summary># Build from source script
curl -fsSL https://raw.githubusercontent.com/sh3ll3x3c/native-devtools-mcp/master/scripts/build-from-source.sh | bash
Or manually:
git clone https://github.com/sh3ll3x3c/native-devtools-mcp
cd native-devtools-mcp
cargo build --release
# Binary: ./target/release/native-devtools-mcp
</details>
After installation, you can proceed to configure the MCP client to connect to native-devtools as described in the manual/configuration sections of the README.
Additional notes
Tips and common considerations:
- If you use Claude Desktop on macOS, you may need to install the app bundle manually and run the setup to generate the proper configuration. Gatekeeper may require you to open the app from /Applications after download.
- For CI or isolated environments, ensure permissions for Accessibility and Screen Recording are granted where required by the OS.
- The server emphasizes local execution with no data exfiltration by default; network access occurs only when explicitly invoked by an MCP client or for verification of binaries.
- For manual configuration, you can place the server reference in your client’s claude_desktop_config.json or equivalent MCP config with the command + args shown above.
- If you plan to use Android support via ADB, ensure ADB is installed and devices are connected; the server can drive UI actions on connected Android devices as part of the MCP capabilities.
Related MCP Servers
awesome-claude-skills
A curated list of awesome Claude Skills, resources, and tools for customizing Claude AI workflows
Overture
Overture is an open-source, locally running web interface delivered as an MCP (Model Context Protocol) server that visually maps out the execution plan of any AI coding agent as an interactive flowchart/graph before the agent begins writing code.
claude-talk-to-figma
A Model Context Protocol (MCP) that allows Claude Desktop and other AI tools (Claude Code, Cursor, Antigravity, etc.) to read, analyze, and modify Figma designs
flutter-skill
AI-powered E2E testing for 10 platforms. 253 MCP tools. Zero config. Works with Claude, Cursor, Windsurf, Copilot. Test Flutter, React Native, iOS, Android, Web, Electron, Tauri, KMP, .NET MAUI — all from natural language.
mcp-tasks
A comprehensive and efficient MCP server for task management with multi-format support (Markdown, JSON, YAML)
adb
An MCP (Model Context Protocol) server for interacting with Android devices through ADB in TypeScript.