MCP Desktop DVR

What is MCP Desktop DVR?

MCP Desktop DVR is a production-ready MCP server for desktop video capture and analysis on macOS. It enables Claude to analyze screen activity through a 30-minute rolling buffer, using either cloud-based (OpenAI GPT-4o) or local (Tarsier2-7B) AI vision, with OCR as a final fallback.

How to use MCP Desktop DVR?

Install and build the server with npm install && npm run build, then add it to your Claude Desktop configuration as an MCP server (command: node, args pointing to the built index.js). Optionally set the ANALYZER_PREFERENCE environment variable to control which analyzer is used. Ask Claude to start recording, analyze recent activity, or check capture status using the server’s tools.

Key features of MCP Desktop DVR

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

Use cases of MCP Desktop DVR

• Ask Claude what errors are on your screen without manual inspection.
• Analyze the last 30 seconds of workflow for troubleshooting or review.
• Continuously record desktop activity and retrieve specific time segments.
• Capture a specific application’s window for focused analysis.

FAQ from MCP Desktop DVR

What analyzers does it support?

Three analyzers: OpenAI GPT-4o Vision (primary, cloud, requires API key), Tarsier2-7B (local, private, no API key), and basic OCR text extraction (final fallback). The preference can be set to auto, openai, tarsier, or ocr.

What are the system requirements?

macOS 10.15 or later, Node.js 18 or later, and Screen Recording permission must be granted.

Where does the captured video data live?

Capture files (MP4 and segmented GIFs) are saved locally on your Mac. GIF files are preserved alongside the MP4; only the first segment is uploaded to OpenAI when using the cloud analyzer.

What are the known limits?

The server maintains a 30-minute rolling buffer of desktop activity. Processing time varies: OpenAI ~3–5 seconds, Tarsier ~5–15 seconds for 30-second clips.

How is authentication handled?

For OpenAI analysis, you must provide an OPENAI_API_KEY in the environment configuration. Tarsier and OCR require no external authentication. The server communicates via standard I/O through the MCP protocol.