Youtube Transcripter

Overview

Extract a transcript from a YouTube URL and produce a concise, structured summary focused on key content, learnings, and actionable suggestions.

Workflow

1) Confirm input

Require a YouTube URL or video ID. Multiple inputs are allowed.
Assume no timestamps unless the user explicitly asks.
Default language is auto-detected; ask only if a specific language is required.

2) First-time setup (one-time)

Make sure at least one transcript extractor is installed. Prefer yt-dlp:

choco install yt-dlp -y

If Chocolatey is not available or not elevated, use pip:

python -m pip install yt-dlp

Optional (API-based):

python -m pip install youtube-transcript-api

3) Extract transcript (best-effort with fallback)

Preferred: run the bundled script:

python scripts/get_transcript.py "https://www.youtube.com/watch?v=VIDEO_ID"

Multiple inputs (URLs or IDs) into a folder:

python scripts/get_transcript.py VIDEO_ID_1 "https://youtu.be/VIDEO_ID_2" --out-dir transcripts

Notes:

--out is only for single input.
--out-dir writes transcript_<id>.txt for each input. Default is transcripts/.
--wrap 125 reflows lines to 125 chars; use --wrap 0 to keep raw line breaks.

To force a language:

python scripts/get_transcript.py "https://www.youtube.com/watch?v=VIDEO_ID" --lang en --out transcript_VIDEO_ID.txt

Fallback order if the script fails:

Install yt-dlp (prefer Chocolatey if available)

Check:

choco -v

If Chocolatey is available, install (requires admin shell):

choco install yt-dlp -y

If Chocolatey is not available or not elevated, use pip:

python -m pip install yt-dlp

yt-dlp with auto captions (binary or Python module)

yt-dlp --skip-download --write-auto-sub --sub-lang en --sub-format vtt "https://www.youtube.com/watch?v=VIDEO_ID"

Then strip timestamps and cue markers from the .vtt file.

If tools are unavailable, ask the user for the transcript text or a downloaded caption file.

4) Capture metadata (title + channel)

If yt-dlp is available, fetch title and channel:

yt-dlp --print "%(title)s|%(channel)s|%(uploader)s" "https://www.youtube.com/watch?v=VIDEO_ID"

PowerShell-safe quoting:

yt-dlp --print '%(title)s|%(channel)s|%(uploader)s' "https://www.youtube.com/watch?v=VIDEO_ID"

If tools are unavailable, ask the user for the video title and channel name. If yt-dlp warns about missing JavaScript runtime, note that it may omit some fields; use channel if uploader is missing.

5) Summarize and structure notes

Always save the structured summary to a file for agent-friendly reading, e.g. transcript_summary.md in the working directory. Keep the raw transcript as transcript.txt. Always keep raw transcripts in transcripts/ and save summaries in the working directory. For multiple videos, include the YouTube ID in filenames: transcript_<id>.txt and transcript_summary_<id>.md. Write files as UTF-8 to preserve umlauts and other non-ASCII characters. Do not paste the full summary into chat; confirm file paths only.

Use this output template. Keep it tight and information-dense.

# Title
Channel: <channel name>
URL: <youtube url>
Video ID: <id>

## Summary
- 3-6 bullets capturing the core message

## Key Points
- ...

## Learnings
- ...

## Suggestions / Recommendations
- ...

## Action Items (optional)
- ...

## Open Questions (optional)
- ...

Quality checks

Remove filler, intros, sponsor reads, and repeated segments.
Preserve important claims, stats, or steps.
If advice is speculative, label it as such.
Keep language plain and direct.
If you hit HTTP 429 (rate limit), wait and retry later or switch to a different network/IP; changing language rarely resolves rate limits.

Resources

scripts/

scripts/get_transcript.py extracts transcripts for one or more URLs/IDs via youtube-transcript-api with yt-dlp binary or Python module fallback.