Transcript API for developers
Send a YouTube, TikTok or Instagram link. Get back timestamped text as JSON, from published captions or from the audio when there are none.
curl -X POST \
https://transcriptfetch.com/api/v2/transcripts/video \
-H "Authorization: Bearer tf_live_••••" \
-d '{"video": "https://youtu.be/bNpx7gpSqbY"}'{
"ok": true,
"data": {
"kind": "transcript",
"platform": "youtube",
"video_id": "bNpx7gpSqbY",
"duration": 401,
"source": "captions",
"segments": [
{ "start": 41.0, "text": "I studied hundreds of companies" },
{ "start": 46.2, "text": "Five factors shaped their outcomes" },
{ "start": 52.3, "text": "One factor stood above the rest" },
… 43 more
]
},
"usage": { "credits_spent": 1, "balance": 4812 }
}A real request and response. Keys start with tf_live_ and stay on your server.
Who it's for
Plain REST, real SDKs and one predictable JSON shape. Whether it is a feature in your app or a job in your pipeline.
Add transcripts to a product feature without maintaining a scraper.
const t = await tf.transcripts.video(url);
return Response.json(t.segments);Feed whole channels into search, a model or a data warehouse.
await tf.transcripts.channel("@3blue1brown", { limit: 50 });
await tf.transcripts.batch(videoIds); // up to 50What people build
Six products you can build on transcripts, and the calls behind each one.
Index every episode of a show and let listeners search by what was actually said. Each result jumps to the second it was spoken.
Paste a link, get chapters and notes back in seconds. One call returns the full timed transcript to feed your model.
Load a whole lecture playlist, then answer student questions with a link to the exact moment in the video.
Drop a TikTok, Reel or YouTube link in a channel and the bot replies with what the video says.
Search a topic every week and read what creators across TikTok and YouTube are saying about it.
Watch the channels you follow and email a summary of every new upload, without polling them yourself.
How we compare
Same URLs, sent to both APIs at the same instant, one attempt each. We count how many transcripts come back and how long the full text takes to arrive.
We built and ran these benchmarks. The script, every URL and every raw measurement are public, so you can run it with your own keys.
| Platform | Transcripts returned | Median time to full text |
|---|---|---|
| YouTube | 120 / 120 | |
| TikTok | 118 / 108 | |
| 118 / 96 |
| Platform | Transcripts returned | 90th percentile time |
|---|---|---|
| YouTube | 500 / 463 |
Coverage
Also on every key
Transcripts, search, channels, Shorts, playlists and monitors
POST/api/v2/transcripts/videoPOST/api/v2/transcripts/searchPOST/api/v2/transcripts/channelPOST/api/v2/transcripts/playlistPOST/api/v2/transcripts/channelPOST/api/v2/monitorsFAQ
A successful caption fetch or listing costs 1 credit. AI Fallback Transcription costs 1 credit per started 5 minutes of audio, billed only when the transcript is delivered. Failures are free.
We transcribe the audio instead. The response has the same shape, with source set to "audio". Long files run as a job you can poll for free.
Yes. npm install transcriptfetch for Node and pip install transcriptfetch-sdk for Python. Or call the REST API with any HTTP client.
Keep your key server-side. Put the call in an API route or server function and send the result to your frontend.
Every response includes usage.balance. You can also call tf.me(), which validates the key and reads the balance for free.
Get started
No framework, just the standard library and one router.
Okay so nobody talks about this part of moving cities.
Toss the noodles straight from the pot.
We charged from day one.
Most houseplants die from too much water, not too little.