Coming Soon
YouTube Transcript and Metadata APIBuilt for AI Agents.
Your agent hits the YouTube Data API quota after 100 searches, then gets 403 until midnight Pacific. Run it on a server instead and YouTube flags the datacenter IP. This API pulls transcripts, video metadata and whole channels through one endpoint, at a volume the official YouTube API will not serve. It is a paid service, priced low per call, and it is not live yet. Leave your email and I will tell you the day it opens.
- Built for agent-scale volume
- Whole-channel transcripts
- Low per-call pricing
Why agents fail on YouTube data
Every AI agent that reads YouTube hits the same three walls, in the same order, usually within an hour of leaving the developer laptop. None of them are fixed by writing better code.
The quota wall
The YouTube Data API v3 gives every project 10,000 quota units a day. A single search.list call costs 100 units, so 100 searches exhaust the entire day and every call after that returns 403 quotaExceeded until midnight Pacific. There is no credit card that lifts it. More quota means a manual audit form, a Google review and a wait, and agent-shaped workloads are exactly the profile those reviews reject.
The datacenter IP problem
The scraper that worked on your laptop starts returning 403 the moment it ships. AWS, GCP, Azure and shared VPS ranges are published and pre-classified as datacenter traffic before your first request. YouTube also added a Proof of Origin token requirement in 2024, generated by running JavaScript in a real browser context, which no plain HTTP client can produce.
Transcripts that break on YouTube schedule
The free transcript libraries and MCP servers work beautifully until they do not. Caption endpoints move, tokens expire, languages fall back inconsistently and a channel that returned clean text last week returns nothing today. Keeping that pipeline alive is ongoing work that has nothing to do with the product you are actually building.
Who this is for
This is built for people running YouTube reads at a volume where the free path stops working, not for anyone fetching a handful of videos a week.
- An AI agent that researches, summarises or monitors YouTube content at scale
- An MCP server author whose get_transcript tool keeps failing under real load
- A research pipeline that needs every transcript from a set of channels, not one video
- A media monitoring or competitive analysis tool tracking channels daily
- A dataset team building a corpus from public educational or conference content
- Anyone who has already filled in Google's quota extension form and been turned down
What the API returns
Video transcripts
Captions for a single video by id or URL, with language selection and fallback. Returned as timestamped segments and as cleaned prose, so an agent can quote a moment or summarise the whole thing without post-processing either format itself.
Whole-channel transcripts
Point it at a channel and get transcripts across the uploads, paginated and resumable. This is the part the free tools cannot do, because it is exactly the pattern that trips rate limits and bot detection.
Video and channel metadata
Title, description, publish date, duration, view and like counts, thumbnails, channel id and public tags. Thin factual data, returned in one predictable shape, with no per-field quota arithmetic to plan around.
Search without the ceiling
Keyword search returning video ids and metadata, at a volume where search.list and its 100-unit price tag would have ended your day before lunch.
Failover across upstream sources
Requests are pulled through multiple upstream sources rather than one. When a source rate-limits, changes shape or goes dark, the request routes around it. Staying unblocked is the actual product here, and it is the part nobody wants to maintain themselves.
How it will work
Join the waitlist
Leave your email below. You get one message when the API opens, with a key and the docs. No sequence, no newsletter.
Call one endpoint
A REST endpoint with bearer auth, plus an MCP server so agents can use it as a tool without any glue code. A GET to /api/yt-pulling returns the live index of every paid endpoint and its price.
Let the failover handle YouTube
No proxy pool, no Proof of Origin token pipeline, no cookie rotation and no library that breaks on a Tuesday. Those problems stay on this side of the API.
Pay for what you call
Usage-based pricing, set deliberately low per call and published in full at launch. Paid from the first call, with no free tier that quietly turns into a quota wall of its own.
Waitlist
Get notified when the API opens
One email when it launches, containing a key and the docs. Nothing else, and nothing shared with anyone.
Building an agent right now? The endpoint is already live at /api/yt-pulling and answers 404 with a JSON body telling you how to get notified. Machine-readable from today.
Why me
An API built by someone who already measures agent traffic
This site already runs an endpoint that fingerprints AI crawler visits in production, and an /agents page written to be read by machines rather than people. Ten years of production React and Next.js sits behind that, along with a paid audit service running on the same domain. The API is the next step from a position I have been holding for a while, rather than a first guess at what agents need.Read: Your API Isn't Ready for Agent Customers →
Frequently Asked Questions
How do I get around the YouTube Data API quota limit?
You cannot raise it with money. The 10,000 daily units are fixed per project, and the only official route to more is a manual audit and quota extension form that Google reviews by hand and frequently declines for scraping or bulk-harvest use cases. The practical options are running many projects, which violates the terms, or using a third-party API that absorbs the volume. This API is the second option.
Why does search.list burn my quota so fast?
A single search.list call costs 100 quota units against a daily budget of 10,000, so 100 searches finish your day. By comparison videos.list costs 1 unit. Any agent that searches before it fetches will hit the ceiling long before it hits a useful result count.
Does this API download YouTube videos?
No. It returns transcripts and metadata only. No MP4, no MP3, no audio extraction and nothing that touches YouTube media streams. That boundary is deliberate and it is not going to move.
Can I get transcripts for an entire YouTube channel?
Yes, and that is the main reason this exists. You point it at a channel and get transcripts across its uploads, paginated and resumable. Doing that yourself is what triggers the rate limits and bot detection that free tools fall over on.
Why does my YouTube scraper work locally and fail on the server?
Almost always the egress IP. Cloud provider ranges are published and classified as datacenter traffic before your first request, so the same code that works from a home connection gets challenged from AWS or GCP. The Proof of Origin token requirement, which needs a real browser JavaScript context, is the usual second cause.
Is there an MCP server for this?
Yes, at launch. The same endpoints will be exposed as MCP tools so an agent can call them directly. The free YouTube transcript MCP servers work well at small volume, and this is aimed at the point where they stop.
What does it cost?
Usage-based, and deliberately cheap per call. The whole point of the service is to cost obviously less than maintaining proxies, Proof of Origin tokens and a caption pipeline in-house, so it is priced low enough that the build-versus-buy question answers itself. It is paid from the first call, with no free tier. Full tiers are published at launch, and waitlist subscribers see them first.
When does it launch?
No date announced yet. A GET request to https://dominikgronkiewicz.com/api/yt-pulling returns 404 while it is being built and 200 with the full index of paid endpoints once it is live, so you can poll it instead of trusting a launch date.
Stop maintaining a YouTube scraper
Proxy pools, Proof of Origin tokens, caption endpoints that move and a quota that ends your day at 100 searches. Leave that on my side of the API, for a per-call price well under what maintaining it yourself costs, and get back to the product you are actually building.
Join the Waitlist