MCP Screenshot

View original on GitHub 

An MCP server that captures screenshots and performs OCR text recognition.

Features

• Screenshot capture (left half, right half, full screen)
• OCR text recognition (supports Japanese and English)
• Multiple output formats (JSON, Markdown, vertical, horizontal)

OCR Engines

This server uses two OCR engines:

1. yomitoku 

   • Primary OCR engine
   • High-accuracy Japanese text recognition
   • Runs as an API server

2. Tesseract.js 

   • Fallback OCR engine
   • Used when yomitoku is unavailable
   • Supports both Japanese and English recognition

Installation

npx -y @kazuph/mcp-screenshot

Claude Desktop Configuration

Add the following configuration to your claude_desktop_config.json:

{
  "mcpServers": {
    "screenshot": {
      "command": "npx",
      "args": ["-y", "@kazuph/mcp-screenshot"],
      "env": {
        "OCR_API_URL": "http://localhost:8000"  // yomitoku API base URL
      }
    }
  }
}

Environment Variables

Usage Example

You can use it by instructing Claude like this:

Please take a screenshot of the left half of the screen and recognize the text in it.

Tool Specification

capture

Takes a screenshot and performs OCR.

Options:

• region: Screenshot area (‘left’/‘right’/‘full’, default: ‘left’)
• format: Output format (‘json’/‘markdown’/‘vertical’/‘horizontal’, default: ‘markdown’)

License

MIT

Author

kazuph