Tutorial · curl, Node.js, Python
How to scrape YouTube transcripts with Python, Node.js or curl (2026)
A walkthrough for GET /v1/youtube/videos/transcript, with the same request in curl, Node.js and Python. Copy the snippet, run it, then read the JSON.
GET /v1/youtube/videos/transcript · 1 credit when a transcript is found
In this tutorial
You'll pull a YouTube transcript from a public video URL and see the text you get back.
Typical job: feeding videos into an AI search index. Store data.transcript.plainText for search, and keep the segments when you want to link to the moment something was said (startMs). Watch links, youtu.be links and Shorts all work. This reads the video's captions; it doesn't transcribe audio.
Step 1
Get ready
Create a Social Fetch API key, then get curl, Node.js or Python ready. New accounts include 100 free credits.

API key. Create a Social Fetch account, open API Keys and create a key. New accounts include 100 free credits.
curl. Any shell with curl. Set SOCIALFETCH_API_KEY to your key (export SOCIALFETCH_API_KEY=sfk_…) before you run the sample.
Node.js. Node 18 or later, which has fetch built in. Save the sample as an .mjs file so top-level await works, and set SOCIALFETCH_API_KEY to your key.
Python. Python 3 with requests installed (pip install requests). Set SOCIALFETCH_API_KEY to your key.
When you're done, SOCIALFETCH_API_KEY is set in your shell.
Step 2
Make the request
Pick curl, Node.js or Python, copy the sample and run it. It calls GET /v1/youtube/videos/transcript with your key in the x-api-key header, using this example: /v1/youtube/videos/transcript?url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DaircAruvnKk.
Each sample prints the JSON response with data and meta. On an error status, the Python sample stops at raise_for_status() and the Node.js sample throws with the error body, because it checks response.ok first. curl prints the error JSON; add --fail-with-body to make it exit non-zero.
Step 3
Read the response
The response is JSON with two parts: data holds the transcripts and meta holds creditsCharged and requestId. The sample below is a trimmed copy of a real response; the endpoint page lists every field. Credits are charged on completed lookups, not on failed requests.

Check data.lookupStatus first when the response has one. Keep meta.requestId in case a row looks wrong.
Watch out
there's no AI fallback on this route. A video with no captions comes back with transcript null, and that lookup is free. Keep the segments if you need timestamps.
Step 4
You finished
The Playground, the docs and your code all call the same endpoint at the same price. Pick a next step below.
FAQ
Common questions
Do I need an official YouTube developer account?
No. Social Fetch authenticates with your Social Fetch API key. You call GET /v1/youtube/videos/transcript and we handle the lookup. You still need to follow each platform's terms for how you use public data.
What does the YouTube transcripts endpoint return?
JSON with two parts: data holds the transcripts and meta holds creditsCharged and requestId. The endpoint page and the API docs describe every field.
Should I use plainText or segments from a YouTube transcript?
Use plainText when you want one string for search or an AI model. Use segments when you need each line with its timing (text, startMs, endMs), for citations or a player. Both are in data.transcript when a transcript is found.
Does YouTube transcript lookup run speech-to-text when captions are missing?
No. This route returns the captions a video has. When it has none, transcript is null and you aren't charged; data.availableLanguages lists the caption tracks a video offers. Watch links, youtu.be links and Shorts all work.
How much does a successful request cost?
1 credit when a transcript is found. Free when a transcript is unavailable.
Is this okay for production?
Yes. It's the same API the Playground and the docs use: typed errors, a requestId on every response, and credits charged only on completed lookups. New accounts get 100 free credits.
Any curl-specific tips for this request?
Yes. Pass the key with -H "x-api-key: $SOCIALFETCH_API_KEY", and wrap the full URL in double quotes so the shell doesn't split it at the & between query parameters. -sS hides the progress meter but still shows connection errors. curl doesn't treat an HTTP error status as a failure, so read the error JSON it prints, or add --fail-with-body.
Any Node.js-specific tips for this request?
Yes. Native fetch doesn't throw on 4xx or 5xx responses, so check response.ok (or status) before parsing JSON. Keep meta.requestId from the body in case you need support with a row.
Any Python-specific tips for this request?
Yes. Call response.raise_for_status() before response.json(), set a timeout on requests.get, and keep response.json()["meta"]["requestId"] when a row looks wrong.
Related