For YouTube transcripts inside a RAG pipeline, an agent or a summariser, start with Scavio's YouTube API. POST /api/v1/youtube/transcript takes a video id or URL and returns the whole transcript as plain text or as timed SRT in one synchronous call. In our test it answered all 20 videos from a YouTube search, including a 2-hour-33-minute course that came back as 27,708 tokens of text in 2.6 seconds. The same key lists a channel's uploads, returns the metadata you store next to each chunk and searches YouTube. A transcript costs 5 credits, $10 per 1,000 on the Growth plan billed yearly.
You need a third-party API because YouTube's own does not hand out other people's transcripts. The Data API's captions.download method needs OAuth authorization, "requires the user to have permission to edit the video" and costs 200 quota units a call, so it serves your own channel, not a knowledge base built from anyone else's. The open-source youtube-transcript-api library works on a laptop, but its README warns that YouTube blocks most IPs known to belong to cloud providers.
We build Scavio, so read this with that bias in mind. Every Scavio number below comes from live calls we made on October 27, 2026. Every competitor detail was read off the vendor's own page or documentation that day, and where a detail was not published we say so. The screenshots are each vendor's own page, captured the same day.
Which YouTube transcript API fits your job?
| Tool | Choose it for | Output | Beyond transcripts | Price (checked Oct 27, 2026) |
|---|---|---|---|---|
| Scavio | RAG and agents that also need channel uploads, metadata and search | Plain text or SRT; lists available languages when yours is missing | Search, channel uploads, video metadata, comments, Shorts, plus Google, Reddit and TikTok on one key | 5 credits a transcript: $10 per 1,000 (Growth, yearly); 50 free credits |
| Supadata | Transcripts across several video platforms, with AI transcription when a video has no captions | Timestamped segments or text | AI transcription fallback; batch and translate on Pro and above | 1 credit a transcript: $17/mo for 3,000, $47/mo for 30,000; AI transcription 2 credits a minute; 100 free a month |
| TranscriptAPI | Low cost per transcript with search and channel listing | Text with or without timestamps | Search, channel videos, playlists, MCP | $5/mo for 1,000; top-ups $2.50 per 1,000 ($1.50 on the annual plan); 100 free once |
| SerpApi | Teams already on SerpApi for Google | JSON, HTML or Markdown | Chapters and language details | $25 for 1,000 searches, $10 per 1,000 on Production; 250 free a month |
| Scrapingdog | Bulk caption pulls at the lowest price | Segments with text, start and duration | Other scraping APIs on the same credits | 5 credits a transcript: $40/mo for 200,000 credits, about $1.00 per 1,000; 100 free credits |
| youtube-transcript-api | Local scripts and prototypes | Python objects; JSON, text, SRT and WebVTT formatters | Translation of transcripts | Free (MIT); cloud IPs are blocked, and the README recommends rotating residential proxies |
How big is a YouTube transcript?
For RAG the first question is not whether you can get a transcript but how much text you are about to embed. We searched YouTube for "retrieval augmented generation explained" with Scavio and pulled the transcript of every one of the 20 results, one call each.
Nineteen came back in English. The median was 196 tokens per minute of video (o200k_base tokenizer), ranging from 141 for a measured IBM Technology explainer to 310 for a fast talker. Fetch time did not grow with length: the median call took 2.1 seconds, the 2-hour-33-minute freeCodeCamp course took 2.6 seconds for 27,073 words and 27,708 tokens, and the slowest call, 6.5 seconds, was a 17-minute video. The 20th result, a 17-minute Apna College lecture, had only a Hindi track; the API said so, and a retry with "language": "hi" returned 3,771 words. Across the 19 English videos, 7.2 hours of video became 83,000 tokens, which is why a long video has to be chunked before it goes into a vector store.
1. Scavio: transcripts plus the channel, metadata and search around them
Scavio's YouTube API is 17 POST endpoints. Measured on October 27:
- Transcript: a video id or any YouTube URL in,
contentout, withlanguage_code,language_nameandformat. All 20 test videos answered, in a median 2.1 s. Ask for"format": "srt"and the 6-minute-36-second IBM explainer came back as 92 timed cues, the input you need to cite a timestamp. When the requested language is missing,contentis null andavailable_languageslists the ones that exist. 5 credits. - Channel videos:
@IBMTechnologyreturned 30 uploads a page with video id, title, URL and duration, and a cursor for the next page. 1 credit a page. - Video: title, channel,
published_at,length_seconds,view_count, keywords and the list of caption tracks, the metadata you store next to each chunk. 1 credit. - Search, comments and Shorts: 20 videos per search page for 2 credits, plus comment threads, Shorts, related videos and community posts.
What it does not do: at 5 credits a transcript, it costs more per transcript than Supadata, TranscriptAPI or Scrapingdog, and a call for a language the video lacks is still billed. The benefit is one key and one JSON format for the whole pipeline, from finding the videos to fetching and citing them, alongside Google, Reddit and TikTok, plus an MCP server that gives an agent a transcript tool. How to use YouTube transcripts for RAG walks through each call with captured responses.
2. Supadata: transcripts with an AI fallback
Supadata's YouTube Transcript API returns transcripts with timestamps and multi-language support, and falls back to AI transcription when a video has no captions. Pricing counts 1 credit per transcript and 2 credits per minute of AI-generated transcript, so a 10-minute video without captions costs 20. Plans run from 100 free credits a month to $5 a month billed annually for 300, $17 a month for 3,000 and $47 a month for 30,000 (about $1.57 per 1,000 transcripts), and credits do not roll over. Batch and translate endpoints start on Pro. Choose it when many of your videos have no captions or you also need transcripts from TikTok, Instagram, X or Facebook.
3. TranscriptAPI: cheap transcripts with search and channels
TranscriptAPI sells transcripts with or without timestamps at 1 credit each, and YouTube search, channel search, channel video lists and playlists at 1 credit a page, with handle resolution and new-upload tracking free. It has a REST API and an MCP server. Pricing is $5 a month for 1,000 credits, or $54 a year, with top-ups at $2.50 per 1,000 (monthly) or $1.50 per 1,000 (annual), and 100 free credits once. Choose it if YouTube transcripts are the only data you need and price per transcript decides.
4. SerpApi: a transcript engine next to Google
SerpApi's YouTube Video Transcript API takes a video id v, an optional language_code and type=asr for auto-generated captions, and returns snippets with start and end times, chapters and language details as JSON, HTML or Markdown. If the requested language is missing it uses the first available one. Pricing is per search: 250 free a month, $25 for 1,000, $75 for 5,000 and $150 for 15,000 ($10 per 1,000), and cached results within an hour are free. Choose it if your team already pays for SerpApi.
5. Scrapingdog: the lowest price per caption
Scrapingdog's YouTube Transcripts API takes v, language and country and returns a transcripts array of segments with text, start and duration. Its documentation says each successful request costs 5 credits, and plans start at $40 a month for 200,000 credits, about $1.00 per 1,000 transcripts, with 100 free credits and no card. Choose it for large caption pulls where you do not need channel listing or metadata from the same API.
6. youtube-transcript-api: the free library
The open-source youtube-transcript-api library (MIT, about 8.5k GitHub stars) is what most YouTube RAG tutorials use: YouTubeTranscriptApi().fetch(video_id) returns the transcript, with formatters for JSON, plain text, SRT and WebVTT and a translate() method. It needs no API key. Its README also says YouTube "has started blocking most IPs that are known to belong to cloud providers", that you will likely see RequestBlocked or IpBlocked exceptions on cloud platforms, and recommends rotating residential proxies. Choose it for local scripts and prototypes, and budget for proxies before you deploy it.
Apify actors and other scraping APIs also sell YouTube transcripts; check whether each one bills per video or per minute before you compare.
Pick by the number you need
A RAG index is a batch job followed by a trickle of new uploads. Indexing 20 channels with 500 videos each is 10,000 transcripts, 50,000 credits, plus about 340 channel-list pages at 1 credit each: about $100 on the Growth plan billed yearly, once. At our sample's median of 196 tokens per minute, 10,000 ten-minute videos are about 20 million tokens to embed. Keeping it current at 100 new videos a week is 500 credits a week, about $1. If you need nothing but transcripts at the lowest price, a per-transcript vendor will cost less; if the same pipeline also finds the videos, stores their metadata and pulls comments or Google results, one key is simpler. Try the YouTube API free, 50 credits, no card, or try the endpoints in the playground.
Reference: all 17 YouTube endpoints · transcript docs · how to use YouTube transcripts for RAG · is there a stable way to pull YouTube transcripts?
Frequently asked questions
Does the official YouTube API return transcripts?
Only for videos you can edit. The YouTube Data API's captions.download method needs OAuth authorization, requires the user to have permission to edit the video, and costs 200 quota units a call. To read the transcript of any public video, you need a third-party API or an open-source library.
Is there a free YouTube transcript API?
The open-source youtube-transcript-api Python library is free under the MIT license, but its README warns that YouTube blocks most IPs belonging to cloud providers, so production use usually means paying for proxies. Hosted APIs have free tiers: Scavio gives 50 credits on signup (10 transcripts), Supadata 100 credits a month, TranscriptAPI 100 credits once, Scrapingdog 100 credits and SerpApi 250 searches a month.
How many tokens is a YouTube transcript?
In our sample of 19 English videos on October 27, 2026, the median was 196 tokens per minute of video with the o200k_base tokenizer. A 10-minute explainer came to about 2,000 tokens, a 49-minute lecture to 8,594 and a 2-hour-33-minute course to 27,708, so long videos need chunking before they go into a vector store.
What is the cheapest YouTube transcript API?
Per transcript, Scrapingdog is about $1.00 per 1,000 on its $40 Lite plan, TranscriptAPI $1.50 per 1,000 with annual top-ups and Supadata $1.57 per 1,000 on its $47 Mega plan. Scavio is $10 per 1,000 transcripts on the Growth plan billed yearly, and SerpApi $10 per 1,000 on Production. Scavio costs more per transcript but includes channel listing, metadata, search and other platforms on the same key.
What happens when a video has no transcript in my language?
It depends on the API. Scavio returns content null with an available_languages list and a message such as "No English transcript for this video. Available languages: hi.", so you can retry with one of them; the call is still billed. SerpApi falls back to the first available language, and Supadata can generate a transcript with AI at 2 credits per minute of video.