YouTube Search Results scraper

What it collects

Search & SEO

The videos that rank for a keyword, with views, channel, length and age. Feed it the keywords the Ideas cards found and the table answers the question those cards cannot: is anyone already serving this, and how well. It is a scraper, which means the whole contract is: fill in the form, and one table comes back. No browser window opens, it costs none of your plan’s automation slots, and the requests go out over plain HTTP carrying a real browser’s TLS fingerprint.

It runs fine from your own address; a proxy is available on the run and this target does not need one. Whichever way it went, every row records it — see the provenance columns below.

What you give it — 6

The form the app draws, read out of the card itself. Each hint below is the one shown beside that field in the run dialog, so nothing here is a description of the product written separately from it.

The one answer it needs

queries

Keywords · list · required

One search per line — the same words you would type into YouTube.

Everything else is optional

sort

Sort by · select

Relevance is YouTube’s own ranking, and it is what you want when asking “what would a viewer see”. Most viewed answers “what is the ceiling on this topic”.

country

Country · text

Two-letter country code. YouTube ranks differently per country.

language

Language · text

pages

Pages per keyword · number

About twenty videos a page. Two pages is the first screen and a bit — past three, results drift off the keyword.

minViews

Minimum views · number

Drop everything below this. Useful for finding the ceiling on a topic; leave empty to see the long tail too.

What comes back — 14 columns

One dataset per run — a typed table your workspace owns, which you can then sort, filter, edit in the grid, export whole, or read back over the local API. These are the columns it is created with.

10 from YouTube Search Results

4 that every scraper writes

The same four on every card, so a table can still answer months later how its rows got there: which service they came from, when, which profile’s identity the requests carried, and whether that identity was signed in.

Four ways to run it

The Scrapers tab. Pick the card, fill in the form, press Start. Every run started here makes a new table, named after what you searched for and when. Check first if you like — it collects one page, writes nothing, and reports which columns came back filled.

Ask the assistant. It has the whole catalogue, so this card is a sentence rather than a form. It fills in the parameters above from what you said and shows you them before it starts.

Over MCP or the local API. The same card, from a coding agent — argus_run_scraper over the MCP server, or POST http://127.0.0.1:39219/v1/scrapers/run on the local API. There is a sample call beside it that writes nothing.

{
  "kind": "youtube_videos",
  "inputs": {
    "queries": [
      "…"
    ]
  }
}

As a step inside a workflow. The Run scraper step puts this collector in the middle of an automation — collect, then filter, then mail — and the tree still opens no window doing it. It is also the one caller that can overrule the table-per-run rule: pointed at a table you name, it creates that table with the columns above if it is missing and files every run into it after, appending or updating the row in place on a column you match on. The step hands the next one the table’s name, its id and the row count.

Where it stops

It will not drive a page. Anything that needs a real browser — a site with no list endpoint behind it, or anything operating an account of your own — is an automation instead, and that is a different tool rather than a worse version of this one.

It signs in to nothing. Reading a public page signed out is the settled position — Bright Data scraped Meta signed out and won; hiQ scraped LinkedIn signed in and lost — so this card asks for no account and holds none.