AgentViewport

AgentViewport

Enables AI agents to view and control the local desktop through browser-based viewport and MCP tools.

Category
访问服务器

README

AgentViewport

AgentViewport is a bridge tool that turns your local desktop into a browser-based viewport and an MCP (Model Context Protocol) Server. It enables AI agents (like Claude Desktop, Cursor, or custom agents) to "see" your screen and interact with it (click, type, drag) via a standardized protocol using their embedded browser tools.

<img src="icon.png" width="200" alt="AgentViewport Icon" />

Features

  • MCP Server: Exposes get_screenshot, mouse_click, key_type, and more to AI agents.
  • Low Latency Viewport: Streams your desktop to http://localhost:3000 for remote viewing or debugging.
  • System Tray Integration: Background execution with a convenient tray menu.
  • Safety Mode: Emergency kill-switch (defaults to Tray Exit).
  • Portable: Can be built into a standalone .exe with bundled dependencies.

Prerequisites

  • Node.js 20+ - Download from nodejs.org
  • Windows 10/11 - Currently Windows-only (uses native Windows APIs for screen capture and input simulation)
  • npm - Comes with Node.js

Installation

Option A: Run from Source

  1. Clone the repository:

    git clone https://github.com/guybnd/agent-viewport.git
    cd agent-viewport
    
  2. Install dependencies:

    npm install
    
  3. Run the application:

    npm start
    

Option B: Build Standalone Executable

  1. Install dependencies (if not already done):

    npm install
    
  2. Build the executable:

    npm run build
    
  3. Copy runtime dependencies:

    node copy_assets.js
    
  4. Run from the dist folder:

    .\dist\AgentViewport.exe
    

[!IMPORTANT] The dist folder contains three items that must stay together:

  • AgentViewport.exe - The main executable
  • vendor/ - Native modules required at runtime
  • icon.png - Tray icon

If you move the executable, move the entire dist folder contents together.

Configuration

Using with Claude Desktop (MCP)

To let Claude "see" your computer, add AgentViewport to your claude_desktop_config.json:

Windows Config Location: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "agent-viewport": {
      "command": "node",
      "args": ["C:\\path\\to\\agent-viewport\\server.js"]
    }
  }
}

If using the built .exe:

{
  "mcpServers": {
    "agent-viewport": {
      "command": "C:\\path\\to\\AgentViewport.exe",
      "args": []
    }
  }
}

Application Config

A configuration file is automatically created at %AppData%\AgentViewport\agent-viewport.config.json. You can modify:

  • port: Web server port (Default: 3000)
  • targetWidth: Screenshot resizing width (Default: 2560)
  • fps: Streaming FPS (Default: 10)
  • jpegQuality: Compression quality (Default: 70)

Tools Available to Agents

Tool Name Description Arguments
get_screenshot Returns a base64 encoded JPEG of the main screen. None
list_monitors Returns screen resolution. None
mouse_click Click at x, y. x, y, button (left/right)
mouse_drag Drag from current pos to x, y. x, y
key_type Type text or press keys. text, key

Security & Privacy

[!CAUTION] Use with Caution: Giving an AI agent access to your screen and input controls (mouse/keyboard) is a high-privilege action. Always monitor the agent's activity in real-time. Do not leave the agent unattended while it has control over your session.

  • Local Execution: The server binds to localhost:3000 by default. It is not accessible from the internet unless you explicitly use a tunnel or port forwarding.
  • MCP Security: The MCP server communicates via stdio (Standard Input/Output), which is a local-only transport managed by your AI client (e.g., Claude Desktop).
  • No Cloud Processing: Screenshots and input data are processed entirely on your machine. No data is sent to external servers by this tool.
  • Transparency: This project is open-source. You can inspect server.js to see exactly how screenshots are captured and how mouse/keyboard inputs are handled.

Troubleshooting

Issue Solution
Port 3000 in use Change port in %AppData%\AgentViewport\agent-viewport.config.json or close the conflicting app
Clicks not registering Make sure you're clicking inside the video area, not the letterboxed margins
Build fails with EPERM Close the running AgentViewport.exe (via tray or Task Manager) before rebuilding
Native module errors Delete node_modules and dist, then run npm install and rebuild
Tray icon not appearing Check the system tray overflow area (^ arrow in taskbar)

Acknowledgements

This project is made possible by these incredible open-source libraries:

License

MIT

推荐服务器

Baidu Map

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。

官方
精选
JavaScript
Playwright MCP Server

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。

官方
精选
TypeScript
Magic Component Platform (MCP)

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。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

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。

官方
精选
本地
TypeScript
VeyraX

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。

官方
精选
本地
graphlit-mcp-server

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。

官方
精选
TypeScript
Kagi MCP Server

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。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

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。

官方
精选