YouTube Scraper — Shorts, videos & search (cached)

Three listings from YouTube, one Actor: a channel's Shorts tab, a channel's videos tab, and search results. videoId, title, view count, publish date, duration, thumbnail. No API key, no browser, no quota.

Ask for	Input
Shorts from a channel	channels: ["@MrBeast"] (default)
Regular videos from a channel	channels: ["@MrBeast"], channelTabs: ["videos"]
Both tabs	channelTabs: ["shorts", "videos"]
Search results	searchQueries: ["ai tools 2026"]

What you get

One row per video (or one row per listing, your choice):

Field	Example
videoId	egvLKQe6I4I
url	https://www.youtube.com/shorts/egvLKQe6I4I
title	Don't Pop the Balloon
viewCount	102000000 — parsed to a number
viewCountText	102M views — YouTube's own wording
thumbnail	highest-resolution source YouTube offers
publishedText	11 days ago — videos and search rows only
durationText	12:56 — search rows only
channelName	on search rows, the channel that posted the video
channel, channelUrl, listing	repeated on every row, so no join is needed
fromCache	whether this came from cache or a live fetch

On a search job, channel holds the query you asked for and listing is search.

How it behaves

Measured against live pages on 2026-08-06 from a plain datacenter IP:

• Every listing ships its data inline as ytInitialData — 48 rows per request for Shorts, ~30 for a channel's videos, ~20 for search — with nothing to render. That is why this Actor needs no browser and stays cheap.
• Pagination goes through YouTube's own InnerTube API. One test run pulled 174 Shorts across 2 channels in 5 requests. Verified per type on 2026-08-06: Shorts 48+48, videos 30+30, search 20+16, with zero overlap between page 1 and page 2 in all three.
• Pagination stops when the listing runs out, not when a page counter says so. Asking for 96 from a channel with 78 returns 78 and spends nothing extra.
• The three types genuinely differ under the hood and are handled separately rather than assumed identical: channel videos arrive as lockupViewModel, search as videoRenderer, Shorts as shortsLockupViewModel — and a search continuation must be POSTed to /youtubei/v1/search, not /browse, which answers a bare HTTP 400 that looks like an expired token.

Deliberate choices you should know about:

• A channel that fails does not fail the run. It comes back as a row with error, so a 20-channel job returns the other 19. Verified: a nonexistent handle produced one error row and the run still exited clean.
• YouTube does not answer overload with a clean error. It serves a consent or challenge page that is still HTTP 200. So a missing ytInitialData is treated as "we were blocked" — never as "this channel has no Shorts". Those two are very different answers and conflating them is how scrapers quietly return empty results.
• Requests are paced under ~1/s and results are cached. A channel's Shorts list changes a few times a week, so a cached answer from an hour ago is the same answer — and it costs YouTube nothing, which is what keeps this Actor off the throttle.
• viewCount is null, never 0, when YouTube's text cannot be parsed. A zero is a lie you could aggregate without noticing.

Limits — read before buying

• View counts are rounded by YouTube itself (102M views), so viewCount is precise to about 1%, not exact. viewCountText is preserved so you can see what was actually said.
• Shorts only. The /videos and /streams tabs use different structures and are out of scope.
• No like counts, comments or publish dates on the channel listing — YouTube does not put them there. Getting those means one request per video, which is a different (and far more expensive) Actor.
• YouTube's Terms of Service restrict automated access. You are responsible for how you use the output. Only public channel listing data is collected.

Input

{
  "channels": ["@MrBeast", "UCHnyfMqiRRG1u-2MsSQLbXA", "https://www.youtube.com/@veritasium"],
  "maxShortsPerChannel": 96,
  "cacheMaxAgeMinutes": 180
}

Handles, bare names, UC… ids and full URLs all work — with or without a /shorts or /videos suffix already on the end.

Development

pip install -r requirements.txt
mkdir -p storage/key_value_stores/default
echo '{"channels":["@MrBeast"],"useApifyProxy":false}' > storage/key_value_stores/default/INPUT.json
python -m src

src/shorts.py is plain standard library and runs without the Apify SDK — useful for checking whether YouTube renamed something:

$python -c "from src.shorts import ShortsClient; c=ShortsClient(); s,t,i=c.first_page('@MrBeast'); print(len(s), i)"