Toucora RAG and AI agents
Feed clean, timestamped transcripts into embeddings, vector databases and LLM pipelines without scraping captions yourself.
Developer infrastructure
The YouTube transcript API developers actually want
One REST API for structured transcripts from public YouTube videos, playlists and channels. Predictable credits, machine-readable output, and clear errors. Free to start.
Request
# quickstart.sh curl "https://api.example.com/v1/transcripts/VIDEO_ID" \ -H "Authorization: Bearer YOUR_API_KEY"
200 OK
{
"video_id": "dQw4w9WgXcQ",
"language": "en",
"duration": 213,
"segments": [
{ "text": "...", "start": 18.64, "duration": 3.24 }
]
}
Every format you need
Ask for the format you already use. Responses are stable and timestamped, so downstream tools never guess.
Transcripts
Single video, batch up to 50, playlists and channels. JSON, TXT, SRT, VTT, Markdown.
GET /v1/transcripts/{videoId}
Jobs & webhooks
Large requests become asynchronous jobs with per-item status, retries and webhooks.
POST /v1/jobs
AI endpoints
Grounded answers with timestamps, summaries and chapters. Small and focused at launch.
POST /v1/ask
Built for real transcript workloads
Research and analytics
Pull transcripts for a playlist or an entire channel in one job, then analyse themes, sentiment or speaking time.
Captions and accessibility
Export SRT, VTT or plain text to publish subtitles, build an accessible archive, or repurpose video into articles.
Automation and tooling
Wire transcripts into MCP servers, Claude Code and Codex so your agents can read any video on demand.
Pricing
Limits are configuration-driven and may change as the API stabilises.
Authentication
Authorization: Bearer YOUR_API_KEY
# or
X-API-Key: YOUR_API_KEY
Keys are hashed at rest, shown once, revocable, and scoped per account.
Errors you can handle
{
"error": {
"code": "TRANSCRIPT_NOT_AVAILABLE",
"message": "...",
"request_id": "req_123"
}
}
Every request carries an X-Request-ID for support and debugging.
OpenAPI
The canonical schema is published and importable into Postman and other tooling.
Frequently asked questions
Is there a free YouTube transcript API?
Yes. Every account gets 10,000 free credits and a developer key — no credit card required. Single-video requests cost one credit each, so you can build and test without paying.
How do I get a transcript from YouTube with an API?
Call GET /v1/transcripts/{videoId} with your key in the Authorization header. You get the transcript, its language, duration and timestamped segments as JSON.
Does the API support captions in other languages?
Yes. List a video’s caption tracks with GET /v1/transcripts/{videoId}/languages, then pass ?lang= to fetch the one you want. Automatic captions are labelled so you know their quality.
What output formats does the API return?
JSON, TXT, SRT, VTT and Markdown. Use ?format= on transcript requests, or convert the JSON yourself — the segments are plain text plus start and duration.
Is there a Python example?
Yes. A few lines with requests or httpx are enough — see the blog guide for a copy-paste Python quickstart, including batching and retries.
Can I get a whole playlist or channel?
Yes. Batch up to 50 video IDs with POST /v1/transcripts, or start a playlist or channel job that returns one result per video with per-item status.
Is there an MCP server for YouTube transcripts?
Yes. An MCP server exposes transcript tools to Claude Code, Codex and other MCP clients, so an agent can pull a video’s transcript without bespoke glue code.
How much does it cost?
The API is free while in early access: 10,000 credits per account and 60 requests/minute. Paid plans with higher limits are coming soon.
Start building with transcripts today
Create a free developer key, make your first request in a minute, and ship transcripts your users can actually search, quote and export.