youtube-transcript-ingestion
breakthrough-edu/youtube-transcript-ingestion
A Claude Code skill that turns any YouTube channel into clean, searchable transcript files -- with built-in rate-limit handling.
Instalação
npx convez-ai breakthrough-edu/youtube-transcript-ingestion1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
YouTube Transcript Ingestion
A Claude Code skill that batch-grabs transcripts from any YouTube channel and saves them as clean, readable markdown files -- so you can research a creator's body of work, build a corpus, or filter content overload without watching every video.
Install
npx skills add breakthrough-edu/youtube-transcript-ingestion
That's it. Update later with:
npx skills update youtube-transcript-ingestion
Manual install (Claude Code only): copy
SKILL.mdinto~/.claude/skills/youtube-transcript-ingestion/.
Requirements
Two command-line tools:
brew install yt-dlp # video metadata (no download)
pip install 'markitdown[youtube-transcription]' # transcript extraction (keep the quotes)
Note: a bare
pip install markitdownwill not fetch YouTube transcripts -- the[youtube-transcription]extra is required.
The skill runs a pre-flight check and tells you exactly what to install if anything is missing.
How to use
In any Claude Code session, just say:
grab YouTube transcripts
The skill then walks you through five steps:
- Channel -- paste a channel URL or
@handle - Filter -- keywords to match in titles + a date range
- Output folder -- where the
.mdfiles land - Context folder (optional) -- point it at your existing notes, and it surfaces only what's genuinely new to you
- Rate-limit strategy -- pacing (a safe delay) or browser cookies (fast, and the only thing that works if you're already blocked)
It confirms the matched video list with you before fetching anything, then saves one markdown file per video with clean YAML frontmatter (title, date, URL, source).
What you get
One markdown file per video, named YYYY-MM-DD-slug.md, with clean frontmatter and the full transcript as readable prose:
---
title: "The video's title"
date: 2026-02-04
url: "https://www.youtube.com/watch?v=..."
source: "ChannelName"
type: video-transcript
transcript_available: true
---
Drop the folder into Obsidian, grep it, or feed it to another AI tool -- it's just text.
When YouTube blocks you
YouTube throttles anonymous transcript requests after ~15-20 from one IP. You'll see HTTP 429 and "Sign in to confirm you're not a bot." The block is sticky -- hours, sometimes a day or more.
The skill handles this two ways:
- Pacing -- a 5-second delay between requests. Fine for small batches, no setup.
- Cookies -- it reads your logged-in browser session via
yt-dlp --cookies-from-browser, which bypasses the limit entirely. No extension, no manual export; just tell it which browser you're signed into YouTube on.
If you're already blocked, only the cookie path works -- waiting and pacing won't help once the block is active.
Why it exists
Learning from a fast-moving channel usually means drowning in hours of video. This pulls the text, filters to the topics you care about, and -- if you give it your notes -- tells you which videos are actually new signal versus stuff you already know.
License
MIT -- see LICENSE. Use it, fork it, share it.