YouTube Comments Scraper: Comments & Replies, No API Key

YouTube Comments Scraper

YouTube Comments Scraper extracts comments and, optionally, their replies from YouTube videos, Shorts and live stream recordings. Each comment is one flat JSON item with text, author, author channel, like count, reply count, creator heart, pinned status and timestamps, plus the video title and channel. It costs $1 per 1,000 comments, replies included.

No YouTube Data API key is needed and there is no daily quota: the actor reads the same comment feed the YouTube website uses.

Features

• Export all comments from a YouTube video to JSON, CSV or Excel. Set maxCommentsPerVideo to 0 for no cap.
• Scrape YouTube comment replies: each reply carries parentCommentId, so threads can be rebuilt.
• Sort YouTube comments by top or newest with sortBy.
• Scrape comments from YouTube Shorts and live stream recordings, not only regular videos.
• Scrape comments for a list of YouTube videos in one run, several videos in parallel.
• Monitor new comments on a YouTube video: with onlyNewComments a scheduled run returns and charges only comments and replies it has not delivered before.
• Flags for pinned comments, comments hearted by the creator, edited comments, verified authors and replies from the channel owner.
• Accepts watch URLs, youtu.be links, Shorts, live and embed URLs, or bare 11-character video IDs.

Use cases

• Sentiment and audience research: what viewers say about a product, launch, creator or ad, exported as a dataset for NLP or an LLM.
• Community management: collect the questions viewers ask so they can be answered or turned into an FAQ.
• Moderation: pull the newest comments and replies to review spam and scam replies.
• Competitor monitoring: schedule a daily run with sortBy: "newest" on a competitor's latest videos.
• Research datasets: comments with like counts and timestamps for topic clustering or classification.

Input

{
  "videoUrls": [
    "https://www.youtube.com/watch?v=aircAruvnKk",
    "https://www.youtube.com/shorts/dQw4w9WgXcQ"
  ],
  "maxCommentsPerVideo": 500,
  "sortBy": "top",
  "includeReplies": true,
  "maxRepliesPerComment": 20,
  "includeVideoInfo": true
}

Field	Type	Default	What it does
videoUrls	array	required	Video URLs or IDs, one per line.
maxCommentsPerVideo	integer	100	Top-level comments per video; 0 means all. Replies are counted separately.
sortBy	string	top	top (YouTube's ranking) or newest.
includeReplies	boolean	false	Also fetch the replies of each comment that has replies.
maxRepliesPerComment	integer	50	Cap per thread; 0 means all replies.
includeVideoInfo	boolean	true	Add videoTitle, channelName and channelUrl to every comment (one small extra request per video).
concurrency	integer	3	Videos processed in parallel, 1 to 10.
proxyConfiguration	object	no proxy	Optional Apify Proxy settings.
onlyNewComments	boolean	false	Monitor mode: output and charge only comments and replies not delivered on an earlier run.
monitorStoreName	string	empty	Key-value store that remembers delivered comments. Empty means a name derived from the videos and options.

Monitor new comments on a YouTube video

Turn on onlyNewComments with sortBy: "newest" to get only the comments posted since your last run:

{ "videoUrls": ["https://www.youtube.com/watch?v=aircAruvnKk"], "sortBy": "newest", "maxCommentsPerVideo": 500, "includeReplies": true, "onlyNewComments": true }

The actor remembers the ID of every comment and reply it delivers in a named key-value store in your Apify account. On the next run those are neither output nor charged. The first run has no history yet: it returns every comment up to the limits and fills the store.

How to set up new-comment alerts (a ready task is on the example page Monitor new comments on a YouTube video):

1. Save the input above as an Apify task.
2. Add an Apify schedule for the task, daily or weekly.
3. Add an integration on the task (email, Slack, Google Sheets, Make, Zapier or a webhook) that fires when a run succeeds. Each run's dataset then holds only the new comments.

Details:

• With sortBy: "newest" a run stops reading a video at the first known comment that is not pinned. With top it reads every page up to maxCommentsPerVideo and drops known comments.
• Replies are checked only under new comments, so a new reply under an older comment is not picked up.
• Without monitorStoreName each different input keeps its own history; the run log names the store.
• Schedule one run per store; two parallel runs can both deliver the same new comment.

Output

{
  "commentId": "Ugwedfi--9yZVC0eefl4AaABAg",
  "videoId": "aircAruvnKk",
  "text": "I study mathematics, physics and architecture. …",
  "author": "@kummer45",
  "authorChannelId": "UCqArZkt3NAOU0tcgIi6xmFQ",
  "authorUrl": "https://www.youtube.com/@kummer45",
  "authorThumbnail": "https://yt3.ggpht.com/…",
  "authorIsVerified": false,
  "authorIsChannelOwner": false,
  "publishedTimeText": "7 years ago (edited)",
  "publishedAt": "2019-08-21T14:03:23.423Z",
  "isEdited": true,
  "likeCount": 13000,
  "replyCount": 166,
  "isHearted": false,
  "isPinned": false,
  "isReply": false,
  "parentCommentId": null,
  "url": "https://www.youtube.com/watch?v=aircAruvnKk&lc=Ugwedfi--9yZVC0eefl4AaABAg",
  "videoTitle": "But what is a neural network? | Deep learning chapter 1",
  "channelName": "3Blue1Brown",
  "channelUrl": "https://www.youtube.com/@3blue1brown"
}

Notes on the data:

• publishedAt is computed from YouTube's relative time ("7 years ago"), so it is approximate. publishedTimeText keeps the original text.
• likeCount is parsed from YouTube's display value, so large counts are rounded the way YouTube shows them (13K becomes 13000).
• Replies have isReply: true and the parent's ID in parentCommentId.
• A video with comments turned off produces one item with error: "COMMENTS_DISABLED". A video that fails for another reason produces an item with the error message.

Pricing

Pay per event: $0.001 per comment (comment-scraped), which is $1 per 1,000. Top-level comments and replies are priced the same. Error items, such as videos with comments turned off, are not charged.

Worked example: 20 videos with maxCommentsPerVideo: 500 and replies off return at most 10,000 comments, so the run costs at most $10. With includeReplies on and maxRepliesPerComment: 20, a video whose 500 top comments have 1,500 replies in total (within the per thread cap) returns 2,000 items and costs $2.

In monitor mode only new comments and replies are charged. A daily run that finds 12 new comments and 3 new replies costs 15 x $0.001 = $0.015. There is no per-run event, so a run with nothing new costs nothing.

FAQ

Do I need a YouTube API key? No. The Data API limits comment reads through its daily quota; this actor reads the public comment feed instead.

How many comments can I export from one video? Set maxCommentsPerVideo to 0 to take everything YouTube serves. Large videos take a while. There are internal safety limits of 5,000 pages of top-level comments per video and 200 pages of replies per thread.

Do I need proxies? Usually not. If requests keep failing and you have not configured a proxy, the actor retries through Apify residential proxy.

What can it not do? It cannot read comments on private or removed videos, or on videos where comments are turned off. It does not return comments from YouTube community posts, and it does not take channel URLs; pass video URLs. To get all video URLs of a channel, run the YouTube Channel Videos Scraper first.

How do I export or schedule it? Download the dataset as JSON, CSV or Excel, call the actor through the Apify API or SDKs, schedule runs, or connect it to Google Sheets, Make, Zapier or n8n. It is also usable as an MCP tool by AI agents.

Use it from Claude, ChatGPT or Cursor (Apify MCP server)

The actor can be called as a tool by any AI assistant that speaks the Model Context Protocol. Add the Apify MCP server to your client with this URL, which loads only this actor as a tool:

https://mcp.apify.com?tools=pulsedata/youtube-comments-scraper

You sign in to Apify on first connection, then ask in plain language, for example "collect the 1,000 top comments on this video and list the most common complaints". Billing is the same as a normal run.

Paying from an AI agent without an Apify account (x402 and Skyfire)

An agent can buy a prepaid, spend-capped Apify token at Apify AGI, paid in USDC through x402 or from a Skyfire wallet, and use it for both the API and the MCP server. This actor runs with limited permissions, so it works with these tokens. The smallest token is $1, tokens expire 14 days after purchase, and Apify marks agentic payments as experimental.

Related actors

• YouTube Scraper: find the videos to analyse through search, channels or playlists, with views and likes.
• YouTube Channel Videos Scraper: list every video of a channel, then feed the URLs here.
• YouTube Transcript Scraper: what was said in the video, to set against what viewers wrote.
• Google Play Reviews Scraper: user feedback on an app, for the same kind of sentiment analysis.
• App Store Reviews Scraper: the iOS side of app feedback.