Youtube Transcript

YouTube Transcript

Fetch auto-generated or manually-uploaded captions from a YouTube video via the youtube-transcript-api library (run via uv run --script).

Usage

# Plain transcript to stdout
scripts/get_transcript.py ""

# With timestamps
scripts/get_transcript.py "" --timestamps

# Specific language (falls back to default if unavailable)
scripts/get_transcript.py "" --lang en

# Write directly to a file
scripts/get_transcript.py "" -o transcript.txt

If the user doesn't specify a filename, either print to stdout or save as -transcript.txt using -o.

Supported URL Formats

• https://www.youtube.com/watch?v=VIDEO_ID
• https://youtu.be/VIDEO_ID
• https://youtube.com/embed/VIDEO_ID
• https://youtube.com/shorts/VIDEO_ID
• Raw 11-character video ID

Output Format

• Plain (default): one line per caption segment.
• Timestamped (--timestamps): [MM:SS] text (or [HH:MM:SS] for long videos).

Handling the Transcript

• Never modify the returned transcript content. Do not paraphrase, summarize, or correct. The raw transcript is the source of truth.
• For plain output, you MAY reflow lines into paragraphs so sentences aren't cut mid-clause. This is cosmetic only — do not change words.
• For timestamped output, preserve the one-segment-per-line structure; do not reflow.

When Captions Aren't Available

The script errors out when a video has no captions at all. Pass that error back to the user — do not fabricate a transcript.

Prerequisites

• uv — installs the Python dependencies in an ephemeral venv at invocation time. Install with curl -LsSf https://astral.sh/uv/install.sh | sh or pip install uv.
• Network access to youtube.com (the script calls the public timedtext endpoint directly).

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: devonjones
• Source: devonjones/devon-claude-skills
• License: MIT

Install and usage instructions live in the source repository linked above.