How to scrape TikTok videos, comments, and profiles

Which TikTok URLs and modes does the scraper accept?

On TikTok the actor runs five of its seven modes: video_detail, user_posts, user_profile, video_comments, and user_likes. Each mode expects a specific URL shape, and the platform field can stay on auto because the actor detects tiktok.com, douyin.com, and bilibili.com hosts by itself.

Video modes take the canonical form www.tiktok.com/@handle/video/1234567890, and profile modes take www.tiktok.com/@handle. Short links copied from the mobile app, vm.tiktok.com and vt.tiktok.com, are followed to their canonical video URL before scraping, so you can paste share links straight from a phone.

There is no keyword or hashtag search input. Hashtags arrive as an array parsed from every caption, so the practical pattern is to scrape creators or videos you already know and filter on that array. The trending mode exists, but it serves Douyin hot search and Bilibili popular feeds, not TikTok.

• Single url or a urls array; the two are merged if both are set, and 500 URLs per run is the recommended ceiling
• maxItems caps paginated modes (user_posts, video_comments, user_likes) at 1 to 5,000 per input URL, default 100
• vm.tiktok.com and vt.tiktok.com share links resolve automatically to www.tiktok.com/@handle/video/<id>
• Unsupported combinations such as live_info or trending on TikTok return a structured error row instead of failing the run
• includeComments: true attaches the first page of comments to a video_detail run without a second call

What fields come back for each TikTok video, profile, and comment?

Every row shares one schema across the three platforms, which means a TikTok video and a Douyin video have identical field names and can land in the same table. The platform and item_type fields tell you what each row is: video, user, comment, live, trending, or error.

Video rows carry title and description, author_id, author_name and author_username, duration_sec, cover_url, and video_url_nowm for the no-watermark MP4. Engagement comes as play_count, like_count, comment_count, share_count, and collect_count, with created_at as an ISO timestamp and music_title plus music_author for the sound.

Profile rows add follower_count, following_count, video_count, bio, avatar_url, and a verified flag. Comment rows carry comment_id, reply_to_id, and comment_text so threads can be rebuilt. Each row also includes raw, the full source response, and scraped_at for freshness tracking.

• Engagement: play_count, like_count, comment_count, share_count, collect_count
• Creator: author_username, follower_count, following_count, video_count, verified, bio
• Assets: cover_url, video_url_nowm (no-watermark MP4, expires within hours), duration_sec
• Sound: music_title and music_author for tracking which audio is spreading
• Threads: comment_id and reply_to_id link replies to their parent comment
• raw keeps the complete source payload for fields the unified schema does not map

How much does it cost to scrape TikTok compared with other scrapers?

Datapika bills per delivered row. A video row costs $0.001, so 1,000 TikTok videos cost $1.00 and a full 5,000-item user_posts pull costs $5.00 (Apify Store, August 2026). Comments and profiles are separate pay-per-event items listed on the store page. There is no subscription, no API key fee, and no charge for URLs that come back as error rows.

The category leader on Apify, clockworks/tiktok-scraper, lists $1.70 per 1,000 results, 245,964 users, and a 4.79 rating as of August 2026 (apify.com/clockworks/tiktok-scraper). It wins on adoption and on input types, since it accepts hashtags and search queries, which Datapika does not. Datapika wins on price per video and on covering Douyin and Bilibili in the same run.

To cap spend, set ACTOR_MAX_TOTAL_CHARGE_USD on the run. When the cap is reached mid-batch, remaining URLs are skipped and written as error rows so the dataset still explains what happened.

• $0.001 per video row, the same price on TikTok, Douyin, and Bilibili
• 1,000 videos for $1.00; 5,000 videos, the maxItems ceiling for one URL, for $5.00
• Error rows (deleted, private, region-blocked, unsupported mode) are never billed
• Comparison: $1.70 per 1,000 results on clockworks/tiktok-scraper, verified August 2026
• Spending cap via ACTOR_MAX_TOTAL_CHARGE_USD; the run stops charging at the limit

How do marketing and research teams use TikTok data from this actor?

The Western marketing case is creator work: vetting influencers, measuring sponsored posts, and spotting sounds before they peak. A user_profile call returns follower_count, video_count, and verified for a shortlist of creators, and a user_posts call on the same handles returns the last 100 posts with per-video engagement, enough to compute average views per post before signing a contract.

For campaign measurement, run video_detail on the sponsored video URLs daily and store play_count, like_count, share_count, and collect_count by scraped_at. The deltas give a growth curve per post without asking the creator for screenshots.

Sound sourcing uses music_title and music_author across a set of trend accounts, and comment mining pairs video_comments with a sentiment model. Because Douyin rows share the schema, a brand active in both markets can watch its Chinese creator partners in the same dataset.

• Influencer vetting: follower_count, video_count, and average play_count from the last 100 posts
• Sponsored post tracking: daily video_detail snapshots keyed by scraped_at
• Sound and hashtag discovery: aggregate music_title and the hashtags array across trend accounts
• Comment mining: video_comments with replies for sentiment and UGC sourcing (requires a TikTok cookie)
• Cross-market benchmarking: TikTok and Douyin rows in one table with identical field names
• AI and ML datasets: captions, stats, and cover images at $1 per 1,000 videos

How do AI agents call the TikTok scraper through MCP?

The actor is exposed through Apify's MCP server, so an agent can discover its input schema and run it without custom glue code. Point the client at https://mcp.apify.com/?tools=fetch-actor-details,openclawai/tiktok-douyin-bilibili-scraper, call fetch-actor-details once to read the mode and URL fields, then call the actor with mode, url or urls, and maxItems.

For plain HTTP, the Apify REST endpoint run-sync-get-dataset-items returns the dataset as a JSON array in one request. Each element is one row with item_type set, so an agent can branch on video, user, comment, or error without parsing HTML.

Bulk runs process URLs concurrently, and 100 URLs typically finish in 2 to 3 minutes, so an agent asked to compare 20 creators gets an answer in one tool call. Comments need a TikTok browser cookie passed in the cookie field; videos and profiles do not.

• MCP endpoint: https://mcp.apify.com/?tools=fetch-actor-details,openclawai/tiktok-douyin-bilibili-scraper
• REST: POST the input JSON to the run-sync-get-dataset-items endpoint and receive rows in the response body
• Every row has item_type, so error handling is a field check, not an exception
• Residential proxy is on by default; custom proxies go in proxyConfiguration.proxyUrls
• 100 URLs in roughly 2 to 3 minutes per the actor documentation, August 2026

Questions, answered

Keep reading