Skip to main content
Glama
Screen Vision MCP Server

Screen Vision MCP Server

by TIMBOTGPT

Screen Vision MCP Server

A Model Context Protocol (MCP) server that provides comprehensive screen capture, OCR, and visual understanding capabilities for macOS.

Features

  • capture_fullscreen: Capture the entire screen

  • capture_window: Capture specific application windows

  • capture_region: Capture defined screen regions

  • extract_text_from_screen: OCR text extraction from screenshots

  • find_text_on_screen: Locate text on screen and return coordinates

  • get_window_list: List all open windows with details

  • get_screen_info: Get display and screen information

  • click_at_position: Automated clicking at specific coordinates

  • monitor_screen_region: Monitor regions for changes over time

  • Screenshot resource management and retrieval

Related MCP server: cua-local

Installation

Quick Install

npm install -g screen-vision-mcp

From Source

  1. Clone the repository:

    git clone https://github.com/TIMBOTGPT/screen-vision-mcp.git
    cd screen-vision-mcp
  2. Install dependencies:

    npm install
  3. Test the server:

    npm start

Usage with Claude Desktop

Add this server to your Claude Desktop MCP configuration (claude_desktop_config.json):

{
  "mcpServers": {
    "screen-vision": {
      "command": "npx",
      "args": ["-y", "screen-vision-mcp"],
      "description": "Screen capture and vision analysis"
    }
  }
}

Or if installed locally:

{
  "mcpServers": {
    "screen-vision": {
      "command": "node",
      "args": ["/path/to/screen-vision-mcp/index.js"],
      "description": "Screen capture and vision analysis"
    }
  }
}

Available Tools

capture_fullscreen

Capture the entire screen.

Parameters:

  • save_path (optional): Custom save path for the screenshot

Example:

{
  "name": "capture_fullscreen",
  "arguments": {
    "save_path": "/path/to/save/screenshot.png"
  }
}

capture_window

Capture a specific application window.

Parameters:

  • app_name (required): Name of the application (e.g., "Safari", "Terminal")

  • save_path (optional): Custom save path

Example:

{
  "name": "capture_window",
  "arguments": {
    "app_name": "Safari",
    "save_path": "/path/to/save/window.png"
  }
}

capture_region

Capture a specific region of the screen.

Parameters:

  • x (required): X coordinate

  • y (required): Y coordinate

  • width (required): Width of region

  • height (required): Height of region

  • save_path (optional): Custom save path

extract_text_from_screen

Capture screen and extract text using OCR.

Parameters:

  • region (optional): Specific region to capture

    • x, y, width, height: Region coordinates

find_text_on_screen

Find text on screen and return its location.

Parameters:

  • text (required): Text to search for

  • case_sensitive (optional): Whether search should be case sensitive (default: false)

get_window_list

Get list of all open windows with their positions.

get_screen_info

Get information about available screens/displays.

click_at_position

Click at a specific screen position.

Parameters:

  • x (required): X coordinate

  • y (required): Y coordinate

  • button (optional): Mouse button ('left', 'right', 'middle', default: 'left')

  • double_click (optional): Whether to double-click (default: false)

monitor_screen_region

Monitor a screen region for changes over time.

Parameters:

  • x, y, width, height (required): Region to monitor

  • duration_seconds (optional): How long to monitor (max 30 seconds, default: 5)

  • interval_ms (optional): Check interval in milliseconds (default: 1000)

Requirements

  • macOS (uses native screencapture command)

  • Node.js 16+

  • Claude Desktop with MCP support

  • Screen recording permissions for automation features

Permissions

On first use, macOS may request permissions for:

  • Screen recording

  • Accessibility (for clicking automation)

  • File system access (for saving screenshots)

Grant these permissions in System Preferences > Security & Privacy.

Screenshots Storage

Screenshots are automatically saved to a screenshots/ directory within the server folder. You can:

  • Access screenshots via the resource URI system

  • Specify custom save paths for individual captures

  • View saved screenshots through Claude's resource system

Development

# Install dependencies
npm install

# Start development server
npm run dev

# Run tests
npm test

Advanced Features

OCR Integration

The server includes hooks for macOS Vision framework integration for advanced OCR capabilities. Full OCR requires additional setup with native macOS Vision APIs.

Automation

The clicking and monitoring features enable automation workflows when combined with other MCP servers.

Security

  • All screen captures require explicit permission

  • File system access is controlled by macOS permissions

  • No network access required for core functionality

License

MIT License - see LICENSE file for details

Contributing

  1. Fork the repository

  2. Create a feature branch

  3. Make your changes

  4. Add tests if applicable

  5. Submit a pull request

Support

For issues and questions, please use the GitHub Issues page.

Available Tools

9 tools
capture_fullscreenC

Capture the entire screen

Inputs: save_path.

capture_regionC

Capture a specific region of the screen

Inputs: x (required), y (required), width (required), height (required), +1 more.

capture_windowB

Capture a specific window by app name

Inputs: app_name (required), save_path.

click_at_positionB

Click at a specific screen position

Inputs: x (required), y (required), button, double_click.

extract_text_from_screenC

Capture screen and extract text using OCR

Inputs: region.

find_text_on_screenA

Find text on screen and return its location

Inputs: text (required), case_sensitive.

get_screen_infoA

Get information about available screens/displays

Inputs: no named parameters.

get_window_listA

Get list of all open windows with their positions

Inputs: no named parameters.

monitor_screen_regionC

Monitor a screen region for changes

Inputs: x (required), y (required), width (required), height (required), +2 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.6/5.0

    Scored across 9 tools

    Disambiguation5/5

    Each tool targets a clearly distinct action: capturing a specific target, extracting/finding text, retrieving system info, clicking, or monitoring. Even the capture variants are disambiguated by their target (fullscreen, window, region).

    Naming Consistency5/5

    All tool names follow a consistent snake_case verb_noun pattern, such as capture_window, get_screen_info, and find_text_on_screen. This makes the tool set predictable and easy to navigate.

    Tool Count5/5

    Nine tools is well-scoped for a screen vision and automation server. Each tool covers a meaningful capability without unnecessary overlap or bloat.

    Completeness4/5

    The tool surface covers capture, OCR, text location, screen/window information, clicking, and region monitoring. Minor gaps exist around input actions like typing or scrolling, but core workflows are well supported.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      B
      quality
      D
      maintenance
      Provides a local MCP bridge for AI assistants to control a Mac by observing screen state and performing actions like mouse movement, clicking, typing, and opening URLs.
      12
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables MCP clients to control macOS via accessibility and screen recording, providing tools to list apps, observe UI, click, type, press keys, and scroll.
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      An ultra-fast native MCP server for macOS desktop automation, enabling visual OCR, text-based clicking, window management, and keyboard/mouse control without hijacking the physical cursor.
      28 npm
      1
      MIT