YouTube Transcript Search (yt-qmd)
Fetches YouTube captions using yt-dlp, converts them to timestamped markdown, and executes hybrid retrieval queries across personal video collections.
YouTube Transcripts (yt-qmd) allows Claude to manage and query custom collections of YouTube transcripts. By combining yt-dlp for caption fetching, VTT-to-markdown conversion, and qmd for multi-vector hybrid retrieval (BM25 lexical search, semantic embeddings, and HyDE with reranking), this skill lets you build a personalized knowledge base from videos, channels, and playlists. Claude can extract verbatim quotes, locate exact video timestamps with deep-links, and summarize concepts across indexed podcast episodes or lectures without relying on pre-packaged corpora.
主な機能
01Structures evidence and briefings in standardized YAML summaries with direct playback links
02Executes hybrid multi-modal queries combining BM25 lexical search, dense embeddings, and HyDE reranking
03Converts raw VTT subtitle files into clean markdown with inline video timestamp deep-links
04Extracts verbatim quotes with automated nearest-timestamp detection for accurate citation
05Fetches captions across individual videos, playlists, or entire channels using yt-dlp with flexible scope filters
061 GitHub stars
ユースケース
01Building a searchable personal research corpus from tech conference talks and podcast channels
02Conducting topic-based briefings across dozens of indexed YouTube videos without watching hours of footage
03Extracting exact verbatim quotes with timestamped playback links for research papers and articles