| name | youtube-transcript |
| description | Fetch the spoken transcript (captions) of a YouTube video or Short. Use when the user gives a YouTube URL or video ID and wants the transcript, captions, spoken-content summary, or quotes from a video. Trigger with phrases like "get the transcript", "transcribe this YouTube video", or "what does this video say". |
YouTube Transcript
Fetch the spoken transcript (captions) from a YouTube video or Short. Delegates
to yt-dlp through the plugin's shared access layer, so it auto-wires a JS
runtime + EJS solver, cookies/proxy, and an optional po_token provider.
How to run
python3 ${CLAUDE_PLUGIN_ROOT}/scripts/fetch_transcript.py \
<url-or-id> [--timestamps] [--lang en] [--proxy URL] [--cookies cookies.txt] \
[--no-check-certs] [--out file.txt]
Requirements
yt-dlp (>= 2025.x): pip install yt-dlp- Recommended:
deno JS runtime — required for the EJS challenge solver on
recent yt-dlp (captions can still fail without it).
Access / credentials
From a datacenter IP YouTube bot-gates or 429s. Configure once; the script
resolves flag > env var > auto-detected file and never prints secret values:
| Env var | Effect |
|---|
YT_COOKIES (YT_TRANSCRIPT_COOKIES) | Netscape cookies.txt — clears 429 |
YT_PROXY (YT_TRANSCRIPT_PROXY / HTTPS_PROXY) | residential egress |
YT_NO_CHECK_CERTS | disable cert checks behind a TLS-intercepting proxy |
A cookies.txt in ${CLAUDE_PLUGIN_ROOT} or beside the scripts is auto-detected.
Full setup — deno + EJS + po_token provider + cookies — is in
resources/unblocking-youtube.md.
Why naive approaches break
- Parsing
ytInitialPlayerResponse from HTML returns empty captionTracks. Never do this.
- Default web/android clients return zero tracks. The script forces
player_client=android_vr,tv,web.
timedtext 429s datacenter IPs. The script retries with backoff; for hard
blocks add cookies or a proxy.- Recent yt-dlp needs the EJS solver for the n-challenge — deno alone is no
longer enough. The access layer adds
--remote-components ejs:npm when deno
is present.
Exit codes
0 ok · 2 no captions · 3 rate-limited (add cookies/proxy) · 4 yt-dlp missing · 5 network error.
Follow-up
To clean/segment the result, use the transcript-processor skill. To capture
screenshots at moments in the transcript, use the video-screenshots skill.
