MCP Desktop DVR

✅ PRODUCTION READY

MCP server for desktop video capture and analysis on macOS. Enables Claude to analyze screen activity through a 30-minute rolling buffer.

Features

- 30-minute rolling buffer of desktop activity
- AI-powered visual analysis with OpenAI GPT-4o (cloud) or Tarsier2-7B (local)
- Intelligent fallback from OpenAI → Tarsier → OCR
- Window-specific capture by application
- Precise extraction of time segments
- Production-ready with comprehensive testing

Requirements

- macOS 10.15+
- Node.js 18+
- Screen Recording permission

Quick Setup

npm install && npm run build

Add to Claude Desktop config:

{

  "mcpServers": {

    "desktop-dvr": {

      "command": "node",

      "args": ["/path/to/mcp-desktop-dvr/dist/index.js"],

      "env": {

        "ANALYZER_PREFERENCE": "auto"  // Auto-select best available analyzer

      }

    }

  }

}

Tools

- start-continuous-capture - Begin recording
- analyze-desktop-now - Extract and analyze recent activity
- get-capture-status - Check recording status
- stop-capture - Stop recording
- configure-capture - Update settings

Usage

Ask Claude:
- "Start desktop recording"
- "Analyze the last 30 seconds"
- "What errors are on my screen?"

Video Analysis Options

The analyze-desktop-now tool supports multiple analyzers with different strengths:

☁️ OpenAI GPT-4o Vision (Primary - Cloud)

- Most accurate analysis using GPT-4o via Responses API - Automatic GIF segmentation - creates 10-second segments from videos - Smart file management - preserves all GIF files locally alongside MP4 - Efficient uploads - sends only first segment to OpenAI for analysis - Comprehensive understanding of complex workflows - Requires OpenAI API key and internet connection - Fast processing (~3-5 seconds)

🤖 Tarsier2-7B (Fallback - Local AI)

- Runs locally on your Mac using Metal Performance Shaders - No API keys required - completely private - Excellent visual understanding of UIs, games, and graphics - Moderate processing time (~5-15 seconds for 30-second clips)

📝 OCR Text Extraction (Final Fallback)

- Basic text extraction from screenshots - Limited effectiveness with modern UIs - Always available as final fallback

Configuration

Set environment variables in your MCP config:

"env": {
  "ANALYZER_PREFERENCE": "auto",       // Options: "auto", "openai", "tarsier", "ocr"
  "OPENAI_API_KEY": "sk-...",          // For OpenAI (recommended)
  "OPENAI_MODEL": "gpt-4o"             // OpenAI model (default: gpt-4o)
}

Analyzer Selection Logic

- "auto" (default) - Uses OpenAI if API key set, otherwise Tarsier, then OCR
- "openai" - Forces OpenAI GPT-4o Vision (requires API key)
- "tarsier" - Forces Tarsier2-7B local AI
- "ocr" - Forces basic OCR text extraction

Documentation

- User Guide - Setup and usage instructions
- CLAUDE.md - Development info

License

ISC