kimtaeyoon83/mcp-server-youtube-transcript

What it solves

This project provides a way for AI models (via the Model Context Protocol) to retrieve and analyze the text transcripts of YouTube videos. It solves the problem of AI models being unable to directly access the spoken content of a video, allowing them to summarize, analyze, or answer questions based on the video's captions.

How it works

The server implements two primary tools:

• get_transcript: Extracts captions and subtitles from YouTube videos, Shorts, or video IDs. It supports language selection with automatic fallback, optional timestamps, and a filter to remove sponsorships and ads based on chapter markers.
• analyze_video: An optional tool that uses the TwelveLabs Pegasus video-understanding model to reason over visual content. This allows the AI to understand what is happening on screen even in videos with little or no speech.

Who it’s for

• AI agents and LLM users (e.g., Claude Desktop) who want to integrate YouTube content into their workflows.
• Developers building AI applications that need to access YouTube transcripts and visual analysis.

Highlights

• Supports multiple YouTube URL formats, including Shorts.
• Built-in filtering to strip out promotional content and ads.
• Optional visual analysis via TwelveLabs for videos without speech.
• Language-specific retrieval with automatic fallback mechanisms.
• Zero external dependencies for basic transcript fetching.