YT MetaScribe MCP Client

An MCP (Model Context Protocol) client for YouTube transcript, metadata and comments extraction — with automatic chunking for long videos and adaptive replies handling for comment threads.

🚀 Installation

For Claude Desktop

1. Install via NPX (recommended):

{
  "mcpServers": {
    "yt-metascribe-mcp": {
      "command": "npx",
      "args": ["-y", "yt-metascribe-mcp-client"],
      "env": {
        "YOUTUBE_MCP_API_TOKEN": "your-api-token-here",
        "YOUTUBE_MCP_SERVER_URL": "https://your-server-url.com"
      }
    }
  }
}

2. Add this to your Claude Desktop configuration:

   • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
   • Windows: %APPDATA%\\Claude\\claude_desktop_config.json

3. Restart Claude Desktop

🛠️ Available Tools

1. get_youtube_transcript — Extract video transcript with automatic chunking for long videos
2. get_youtube_metadata — Get comprehensive video metadata
3. get_youtube_combined — Get both transcript and metadata in one call
4. get_youtube_comments — Fetch comments for a video with adaptive replies handling. Defaults to 200 comments sorted by relevance (hard cap 1000). Replies strategy adapts automatically to the video's total comment count — small videos get every reply, medium get the first few per thread, very large videos get top-level only to keep LLM context predictable.

🔧 Configuration

Environment Variables

• YOUTUBE_MCP_API_TOKEN — Your API token for the YT MetaScribe MCP server
• YOUTUBE_MCP_SERVER_URL — Server URL (optional, defaults to public server)

📝 Usage Examples

With Claude Desktop:

• "Get the transcript for this YouTube video: https://youtu.be/VIDEO_ID"
• "What's the metadata for this video: https://www.youtube.com/watch?v=VIDEO_ID"
• "Analyze this YouTube video: https://youtu.be/VIDEO_ID"
• "Fetch the comments for https://youtu.be/VIDEO_ID and summarise the main themes"
• "Show me the newest 100 comments on https://youtu.be/VIDEO_ID" (uses order=time)

🎯 Features

• Automatic chunking for long videos (1–2+ hours)
• Adaptive comment replies — depth scales to the video's comment volume so LLM context stays predictable
• Compact comment format by default — ~3× less JSON overhead per comment for LLM-friendly responses
• Intelligent streaming with progress updates
• Caching support for improved performance
• Error handling with retry logic

🔗 Links

• GitHub Repository: https://github.com/eviipx/yt-metascribe-mcp-server
• Server Documentation: https://github.com/eviipx/yt-metascribe-mcp-server#readme
• Issues: https://github.com/eviipx/yt-metascribe-mcp-server/issues

📄 License

MIT License

🔄 Changelog

v2.3.1

• 📝 Re-publish so the npm package page reflects the get_youtube_comments tool shipped in 2.3.0. No code changes vs 2.3.0.

v2.3.0

• ✨ New get_youtube_comments tool. Fetches comments from a video with adaptive replies handling based on the video's total comment count: small videos get every reply, medium videos get the first few per thread, large videos get top-level only. Default 200 comments, hard cap 1000, sorted by relevance. MCP responses use a compact format to keep LLM context predictable.

v2.2.1

• 🐛 Fix source="youtube:unknown" in wrapper tags when the tool was called with url instead of video_id. The video_id is now extracted from whichever nested field holds it (metadata.video_id, current_chunk.video_id, etc.).

v2.2.0

• 🔒 Prompt-injection mitigation: every tool response is wrapped in <untrusted_content source="youtube:VIDEO_ID">…</untrusted_content> with a header instructing the model to treat the content as data rather than instructions. Defense against adversarial text embedded in video titles, descriptions, or transcripts.

v2.1.0

• ✨ Clean tool names (removed _streamed suffix)
• 🚀 Direct server-client mapping (no abstraction)
• 🔧 Automatic chunking built into all transcript tools
• 📦 NPM package publication

v2.0.0

• 🆕 Streaming support for long videos
• 🔄 Automatic chunking with 30-minute segments
• ⚡ Performance optimizations