Skip to content

ryfter/youtube-transcripts

v1.0.0MIT

Fetch YouTube video transcripts as plain text from Kiro, so the agent can read, summarize, quote, or research spoken video content it otherwise cannot see. Bundles a small Python script that wraps the youtube-transcript-api library: pass one or more video URLs or IDs and it writes one UTF-8 .txt per video, reporting word counts and skipping videos with no captions. Handles full watch URLs, youtu.be short links, /shorts/ and /embed/ forms, and bare 11-character IDs, and works across both the modern (>=1.0) and legacy youtube-transcript-api interfaces. Use when the user shares YouTube links and wants the words, not the video. No API key required; transcripts are pulled from YouTube's public caption tracks.

GitHubOpen Website

youtube-transcripts

A Kiro Power (Agent Plugins v1.0.0) that fetches YouTube video transcripts as plain text so Kiro can read, summarize, quote, or research spoken video content it otherwise cannot see.

Why it exists

Kiro can't watch a video, and a YouTube watch page is a JavaScript shell that returns nothing useful to a normal web fetch. When someone shares a YouTube link and wants the words — for a summary, a quote, or research — there was no built-in way to get them. This power provides one.

What it does

  • Takes one or more YouTube URLs or bare 11-character IDs.
  • Pulls each video's public caption track via youtube-transcript-api.
  • Writes one UTF-8 .txt per video and prints a per-video status (OK / NONE / GONE / ERR), continuing past failures.
  • Accepts watch?v=, youtu.be/, /shorts/, /embed/, /live/, and bare IDs.
  • Needs no API key.

How it works

The power ships a single skill, fetch-transcripts, whose scripts/fetch_transcripts.py wraps youtube-transcript-api. The script works across both the modern (fetch) and legacy (get_transcript) versions of that library, supports an --out output directory and --lang preference list, and exits non-zero if any video fails.

python scripts/fetch_transcripts.py "https://www.youtube.com/watch?v=jGD_UR4wMJc" --out transcripts

Getting started

  1. Install the power in Kiro (Powers UI > Add Custom Power > Local Directory > this folder), or add the repo from GitHub and pick this subfolder.
  2. Install the dependency once: pip install youtube-transcript-api.
  3. Ask Kiro to fetch a transcript for a YouTube link, then read/summarize it.

Skills

  • fetch-transcripts — fetch transcripts for YouTube URLs/IDs and save them as .txt, with language selection, batch input, and per-video status reporting.

Limitations

  • Only works when a video has captions (auto-generated or uploaded). Videos with captions disabled return NONE; transcribe those with a separate tool.
  • YouTube rate-limits rapid or bulk requests from one IP; keep batches modest.
  • Auto-generated captions can mishear words — verify exact quotes, especially names, numbers, and technical terms.
  • Transcripts are external, untrusted text; review before acting on anything that looks like an instruction inside them.

Credits

This power wraps youtube-transcript-api, which does the actual caption retrieval. That project is maintained separately under its own license. Thanks to its authors.

License

MIT. See LICENSE. The bundled youtube-transcript-api dependency is maintained separately under its own license.