MCP Server Youtube Transcript

What kimtaeyoon83/mcp-server-youtube-transcript MCP server does

The kimtaeyoon83/mcp-server-youtube-transcript MCP server exposes YouTube caption retrieval through MCP. Its main tool, get_transcript, accepts a standard YouTube watch URL, a Shorts URL, or a video ID. The response can use a requested language, add time markers, and remove sponsorship or advertising segments identified through chapter markers.

Transcript retrieval works without a TwelveLabs account or API key. The server also includes an optional analyze_video tool for TwelveLabs Pegasus. This separate path reasons about the visual contents of a video rather than only its spoken text, making it suitable for questions about demonstrations, gameplay, b-roll, music videos, or other footage with limited speech.

How it works

Call get_transcript with the required url value. The lang option defaults to English and falls back to another available language when the requested language is unavailable. include_timestamps defaults to false; when enabled, transcript lines include time references such as [0:05]. strip_ads defaults to true, while setting it to false returns the transcript without that filtering.

analyze_video requires both a direct video URL and a TwelveLabs API key. The URL must point to a publicly reachable video file, such as an .mp4, .mov, or .webm resource, including a pre-signed URL. A YouTube watch page is not suitable because it serves HTML rather than a raw video stream. The tool accepts a prompt, a Pegasus model selection, and a maximum token limit.

Setup and configuration

The kimtaeyoon83/mcp-server-youtube-transcript MCP server can run through npx and uses stdio for MCP communication. Claude Desktop can launch it with the package name and no required environment variables:

{
  "mcpServers": {
    "youtube-transcript": {
      "command": "npx",
      "args": ["-y", "@kimtaeyoon83/mcp-server-youtube-transcript"]
    }
  }
}

Set TWELVELABS_API_KEY in the server's environment to enable analyze_video. The transcript tool remains available when this variable is absent. For local development, the repository documents Node.js 18 or newer, npm or yarn, npm install, npm run build, and npm run watch. Tests run with npm test, and npm run inspector starts the documented MCP Inspector workflow.

Tools and capabilities

• get_transcript: retrieves captions from YouTube URLs, Shorts URLs, or IDs.
• Language selection with automatic fallback when the requested language is unavailable.
• Optional timestamps for locating transcript content.
• Optional removal of sponsorships, advertisements, and promotional sections based on chapter markers.
• analyze_video: sends a direct video file URL and prompt to TwelveLabs Pegasus when configured.
• analyze_video options include pegasus1.2 or pegasus1.5, a prompt, and a maximum response length.

Limitations and notes

The kimtaeyoon83/mcp-server-youtube-transcript MCP server depends on captions being available for transcript retrieval. Invalid URLs or IDs, unavailable transcripts, language problems, and network failures are handled as documented error cases. The analysis feature is not a way to submit a YouTube watch URL; it requires a publicly reachable direct video file and the TWELVELABS_API_KEY environment variable. The project is licensed under MIT.