I built an API that gives any AI the ability to learn from YouTube

YouTube is the world's largest knowledge base. Billions of hours of tutorials, lectures, interviews, and technical walkthroughs — and your AI agent can't touch any of it.

It can read text. It can search the web. But it can't watch a video.

I built TubeScribe API to fix that.

The problem

When you're building an AI agent or RAG pipeline, you constantly run into this wall. A user asks "summarize this YouTube video" or "extract the key points from this lecture" — and you have nothing. The transcript might be buried in auto-generated captions, or not exist at all for older videos.

You could build your own solution. Pull the Innertube API, parse the caption XML, handle Whisper fallback for videos with no captions, deal with rate limits and format variations... or you could just make one API call.

One API call. Any YouTube video. Full transcript + AI summary.

curl "https://transcriptapi.dev/transcript?url=https://youtube.com/watch?v=VIDEO_ID&summary=true" \
  -H "X-API-Key: your_key"

Response:

{
  "title": "...",
  "channel": "...",
  "duration": 1247,
  "transcript": [
    { "text": "Welcome back everyone...", "start": 0.0, "duration": 3.2 },
    ...
  ],
  "summary": "This video covers...",
  "wordCount": 4821,
  "cached": false
}

That's it. Full transcript with timestamps, video metadata, and an optional Claude AI summary — all in one response.

How it works

Caption extraction via Innertube — hits YouTube's internal API to pull auto-generated or manual captions when available. Fastest path, no rate limits.
Whisper fallback — for videos with no captions (older content, music, foreign language), it downloads the audio and runs OpenAI Whisper transcription automatically.
Claude AI summary — optionally pass &summary=true and get a structured summary generated by Claude alongside the raw transcript.
Smart caching — transcripts are cached so repeat requests are instant and cheap.

Use cases

RAG pipelines — index YouTube content into your vector DB. Feed transcripts directly into embeddings.
AI research agents — let your agent learn from any video source, not just text.
Content repurposing — transcript → blog post, thread, or newsletter automatically.
EdTech — build study tools, quizzes, and summaries from educational video content.
Podcast processing — most podcasts are on YouTube. Transcribe and analyze at scale.