YouTube Scraper | Videos, Channels, Search & Comments

Extracts YouTube video details, channel video listings, search results, and video comments using YouTube's own public pages and its internal "innertube" API. No login, no API key sign-up, no browser, no proxy.

What you get

• Video details: title, channel, view count, like count, publish date, upload date, category, description, keywords/tags, duration, live-stream flag, thumbnail.
• Channel videos: a channel's videos (title, view count, published time, duration, thumbnail), paginated past the first page via YouTube's own innertube API.
• Search results: matching videos (title, channel, view count, published time, duration), also paginated.
• Comments: top-level comments on a video (author, text, like count, reply count, publish time, pinned flag), also paginated.
• Incremental mode: track one or many channels over time. Each run after the first returns only videos published since the last run — no re-scraping, no re-paying for old videos.

How this works (honestly)

YouTube's own pages embed the same JSON the site's front-end uses to render itself server-side — ytInitialPlayerResponse (video details) and ytInitialData (channel/search/watch-page context) — plus a ytcfg blob carrying an INNERTUBE_API_KEY. That key unlocks YouTube's internal "innertube" API (/youtubei/v1/browse, /next, /search), the same well-documented, widely-used mechanism tools like yt-dlp rely on for pagination. This Actor:

1. Fetches the relevant public page (a video, a channel's /videos tab, or a search results page).
2. Parses the embedded JSON for the first page of results.
3. Follows the page's own continuation token through the innertube API for further pages, anonymously — confirmed working live for channel videos, search results, and comments alike.

No headless browser, no residential proxy, no login wall to fight — because there isn't one. That keeps this fast, reliable, and cheap, and that cost is passed on to you.

One honesty note: only top-level comments are extracted. Expanding every reply thread would mean a separate continuation call per comment, which turns an unpredictable cost into a genuinely unbounded one — so replies are left uncollected, though each comment's replyCount tells you how many exist.

Incremental mode (channels only)

Set mode: "changes". On a channel's first run, it seeds from firstRunSince (default 30 days back). Every run after that resumes from a per-channel high-water mark stored in a key-value store. Channel grids only expose a relative publish time ("12d ago", "2w ago", "1mo ago" — confirmed live), not an exact timestamp, so the mark uses that as an approximate cutoff; a remembered set of video ids is the real safety net that guarantees no video is ever billed twice, even right at the boundary. A run cut short by your budget still advances the mark safely.

Input at a glance

Field	Description
videoUrls	Videos to fetch full details for — a URL or bare 11-character id, one per line.
channels	Channels to list videos from — a handle (@nasa), a full URL, or a UC... id.
searchQueries	Search terms to run, one per line.
commentVideoUrls	Videos to fetch top-level comments for (separate budget from the above).
mode	"all" (default) or "changes" (incremental) — channels only.
maxItems	Budget cap on billed video rows (details + channel + search combined).
maxComments	Budget cap on billed comment rows. Set 0 to disable comments.
maxCommentsPerVideo	Cap on comments pulled from any single video.
maxPagesPerSource	Cap on innertube pagination pages per channel/search/comment video.
firstRunSince	How far back a channel's first incremental run looks (e.g. "30 days").

Billing (pay-per-event)

Event	Price
Actor started	$0.002 — charged once per run, regardless of results
Video scraped	$0.0007 — video details, a channel's video, or a search result
Comment scraped	$0.00018

Real-measured cost on Apify's own infrastructure is dominated by fixed per-run overhead (well under $0.001 even for a large run), so pricing is a flat per-event rate with no hidden per-GB or "with extra fields" tier — every row already includes everything this Actor can extract for it.