An agent skill that gives Claude Code, Codex, Cursor and other AI agents YouTube transcripts, through the TranscriptFetch API.
- Transcripts of any public YouTube video, with timestamps, from its captions or by AI transcription when it has none
- Find videos by search, channel or playlist, with paging
- Many at once: up to 50 transcripts per request, or a whole channel
- Ready for AI: plain text (about two-thirds fewer tokens), Markdown with timestamp links, or chunks sized for embeddings
npx skills add TranscriptFetch/youtube-transcript-skillOr copy the one file into Claude Code yourself:
mkdir -p ~/.claude/skills/youtube-transcript
curl -fsSL https://raw.githubusercontent.com/TranscriptFetch/youtube-transcript-skill/main/skills/youtube-transcript/SKILL.md -o ~/.claude/skills/youtube-transcript/SKILL.mdFor other agents, put skills/youtube-transcript/SKILL.md in that agent's skills folder.
The first time you ask for a transcript, the agent opens a TranscriptFetch page in your browser. Sign in, check the code and click Approve. The agent gets its own API key, which never appears in the chat and which you can revoke on your API Keys page. No key to copy, no password shared with the agent.
- "Summarize this video: https://www.youtube.com/watch?v=..."
- "Get the transcript of this talk with timestamps and pull out every claim about pricing."
- "Collect the last 50 videos from @channel with their transcripts into a knowledge base."
- "Find recent videos about vector databases that have captions and compare what they say."
50 free credits every month, no card. 1 credit per caption transcript and per listing page; AI transcription of a video without captions is 1 credit per started 5 minutes of audio. Failed requests are free. Plans: transcriptfetch.com/pricing.
- TikTok Transcript Skill
- Instagram Transcript Skill
- MCP server for YouTube, TikTok and Instagram transcripts in any MCP client
- API docs
MIT
