YouTube scraper for agents: videos, stats, transcripts

A competitor has a YouTube channel with three hundred videos. You want to know which ones did well, how often they post, what they talk about and what they say in the best ones. Watching is out of the question. A YouTube scraper that lists the videos and their numbers solves half of it, and the transcripts solve the rest.

This post shows how an AI agent gets that data from the looot catalog. It covers the channel video list, per video stats, transcripts, what each costs, and a prompt that turns a channel into a short report.

What does a YouTube scraper return through looot?

A channel videos call returns a list of a channel's videos with their titles, publish dates and view counts. A video detail call returns more about one video. A transcript call returns the spoken text. Each is a catalog endpoint with a price, and several have more than one provider.

Providers differ on whether they take a channel id or a username, and on which fields they return. Search the catalog in plain words, such as "youtube channel videos", and inspect each endpoint before you rely on its output.

For a single transcript without an agent, the free YouTube transcript tool runs the same kind of call from a web page.

How do I get YouTube channel data without writing a scraper?

You tell the agent which channel you care about and let it run the endpoints. It finds the endpoint by task, reads the input fields, checks the price and calls it. No browser, no script and no parsing.

The agent follows this order:

1. Search the catalog for "youtube channel videos" and "youtube video transcript". Searching is free.
2. Inspect the cheapest endpoint for each and read its fields and price.
3. Resolve the channel to the input the endpoint wants, an id or a username.
4. Pull the video list, then transcripts for only the videos you care about.

Set up looot with one line first. A new workspace starts at zero balance and there is no trial credit, so top up before the first run.

What does a YouTube scraper call cost?

You pay per call. A channel videos call returns a page of videos, so a long channel needs several pages. A transcript call covers one video, so fetch transcripts for a short list and not for the whole channel. The table reads the cheapest priced endpoint from the live catalog when the page loads.

What 1,000 lookups cost shows how to turn a per call price into a job price. Decide the list first, then fetch transcripts for the top ten videos, not the other three hundred.

Which YouTube jobs fit an agent?

The ones where numbers sort the videos and text explains them.

• Find the outliers. Rank a channel's videos by views and flag the ones far above the channel's typical video. Those are the topics and titles that worked for that audience.
• Posting pattern. From the dates, the agent can tell how often a channel posts and whether the pace changed. That is a quick read on how serious a competitor is.
• Title and topic patterns. Ask the agent to list the repeated words and formats in the top ten titles. Use it as input for your own titles, not as text to copy.
• Turn talks into drafts. Pull the transcript of one strong video and have the agent outline a post from it, with quotes. The recipe for YouTube transcripts to blog posts lays out the steps.

Treat view counts as a snapshot. They move, and an old video can have a different count next month. Record the call time next to each number.

How do I check what the agent returns?

Open the channel in a browser and compare three rows: the newest video, a mid-list video and the top video by views. If the titles, dates and counts match, the endpoint is returning what you think it is.

Then check the edges. A channel with thousands of videos may need many pages, and the agent may have stopped early. Ask it to report how many videos it fetched and whether the list ended or hit your page limit. A total with no stop reason is a guess.

Transcripts need a second check. Auto captions contain errors, especially names and product terms. Treat a quote from a transcript as approximate until you have listened to the line.

What does a finished channel review tell you?

Here is an invented example, not a real channel. Imagine one with two hundred videos. The agent fetches the list in four pages and reports that the list ended, not that it hit the limit. It sorts by views and finds that ten videos hold most of the total, and that eight of the ten are tutorials with "how to" in the title. The channel's recent videos, mostly opinion pieces, sit far below those tutorials.

That is a useful finding for the price of four calls. It says the audience came for tutorials, and that the channel's newer direction is not working. You can act on it: write the tutorials the channel stopped making, or watch whether the channel returns to them.

The agent did not tell you why the tutorials worked. It cannot. A transcript read of the top three videos gives you hints, such as the length of the intro and how fast the first answer arrives, and then the judgment is yours.

Run it: a prompt for your agent

When the table looks right, ask for transcripts of the top two videos in a second prompt.

What it cannot do

• It reads public data. It does not upload, comment or manage a channel.
• It does not guarantee a transcript. A video without captions has no text to return.
• It does not return private or unlisted videos from other people's channels.
• It does not run on a timer. Tracking a channel as it grows needs scheduled runs, not available yet, so rerun the job by hand.
• Fields differ by provider, and some take a channel id and others a username. Inspect the endpoint first.

For the same method on short video, see the TikTok guide. To browse the YouTube endpoints and their prices, open the public catalog.

Last checked 2026-10-03 against the live catalog. Author: Walid Boulanouar.