TikTok Scraper: Videos, Profiles, Hashtags, Sounds

Scrape public TikTok data without a login, an account, or an API key. Five modes in one actor, all HTTP-only — no browser, no request signing, no cookies.

Mode	Give it	You get back
video	video url, short link, or numeric id	views, likes, comments, shares, caption, upload time, cover, sound — plus the uploader's follower / likes / video counts
profile	username, @handle, or profile url	followers, following, total likes, bio, verified, private flag, avatar
hashtag	fyp or #fyp	the videos TikTok shows on that tag's page, plus the tag's total video count
music	sound url or sound id	the videos using that sound, plus the sound's artist and total video count
oembed	full video url	TikTok's official lightweight record: title, author, thumbnail

Quick start

{
  "mode": "video",
  "targets": [
    "https://www.tiktok.com/@complex/video/7626254334065511711",
    "https://vm.tiktok.com/ZGdxxxxxx/",
    "7626254334065511711"
  ],
  "proxyConfiguration": { "useApifyProxy": true }
}

Every entry form above works in video mode. Short links cost one extra request to resolve; bare ids cost none.

Sample output — video mode

{
  "_mode": "video",
  "_source": "embed-frontity-state",
  "videoId": "7626254334065511711",
  "description": "can't even lie this is fire 🔥 [🎥: @nasa ]",
  "createTimeIso": "2026-04-08T05:21:59Z",
  "playCount": 11000,
  "diggCount": 427,
  "commentCount": 10,
  "shareCount": 14,
  "authorUniqueId": "complex",
  "authorNickname": "COMPLEX",
  "authorVerified": true,
  "authorFollowerCount": 6300000,
  "authorHeartCount": 394000000,
  "authorVideoCount": 11800,
  "musicId": "7301919715965077506",
  "musicName": "euphoria",
  "musicAuthorName": "inertia. & Aurenth",
  "locationCreated": "US",
  "isAd": false,
  "videoUrl": "https://www.tiktok.com/@complex/video/7626254334065511711"
}

The author block is the part worth noticing: a list of video urls doubles as a creator dataset, because TikTok ships the uploader's account stats inside the video payload. No second request per account.

Sample output — music mode

{
  "_mode": "music",
  "playlistType": "music",
  "entityId": "7301919715965077506",
  "musicArtist": "Aurenth",
  "entityVideoCount": 91900,
  "videoId": "7584740654534642957",
  "playCount": 10200000,
  "authorUniqueId": "coastenitaa",
  "videoUrl": "https://www.tiktok.com/@coastenitaa/video/7584740654534642957",
  "_warning": "single_page_surface_embed_playlist_is_not_paginable;playlist_rows_carry_play_count_only_no_digg_or_comment_counts"
}

What this actor does not do, and why

This is the honest part, and it is short on purpose. TikTok's consumer web pages are behind a WAF that answers with a 1.4 KB stub, and its internal JSON APIs require a request signature minted by TikTok's own obfuscated JavaScript. This actor uses only the embed and oEmbed surfaces, which are published for third-party use. That buys reliability and a tiny cost per row, and it costs the following:

• No comments. /api/comment/list/ is signature-gated. You get the comment count, not the comments.
• No search, no trending, no For You. Same gate.
• No "all videos by this user". Listing an account's posts needs the signed /api/post/item_list/. profile mode returns account stats; to collect a creator's videos you need their video urls or ids.
• Hashtag and sound listings are one page — about 6 and 9 videos. The embed's own loader is the signature-gated endpoint, so maxItemsPerTarget cannot produce more. entityVideoCount tells you how many exist in total so you can see the gap rather than guess at it.
• No Top Ads / Creative Center. Those APIs answer {"code":40101,"msg":"no permission"} to anonymous callers.
• Counters are rounded by TikTok, not by this actor: playCount arrives as 11000, not 11043. videoWidth/videoHeight/videoDuration are often null outright -- TikTok omits the videoMeta block on many videos and ships the placeholder 99 on some of the rest.

Full measurements behind each of these are in CRAWLING_METHOD.md.

Reading the output

Every row carries a small envelope so a mixed dataset stays readable:

• _mode — which surface produced the row.
• _input — the target as resolved from your entry.
• _source — embed-frontity-state or oembed-api.
• _index — position within a hashtag/music listing.
• _warning — a ceiling, not a failure (single-page listing, thin listing rows, rounded play counts, profile without a video list).
• _error + _errorDetail — INVALID_INPUT, NOT_FOUND, PRIVATE_ACCOUNT, EMPTY_RESULT, BLOCKED_AFTER_RETRIES, UPSTREAM_SHAPE_CHANGED.

Every input produces at least one row. A username that no longer exists becomes a NOT_FOUND row, not a silent omission — so you can always reconcile your input list against the dataset line for line.

Proxies and pacing

TikTok throttles embed requests per IP and answers with HTTP 503 after a handful of calls. The actor handles this by rotating the TLS fingerprint and the proxy session, then backing off (2s → 4s → 8s) across four attempts.

Rotating the fingerprint alone does not lift the throttle, so:

• More than a few targets: use a proxy. {"useApifyProxy": true} is the default; residential helps most.
• delaySeconds defaults to 2. Lower it only behind rotating residential proxies.
• A run that hits BLOCKED_AFTER_RETRIES on every target ends as a failed run rather than an empty success, so a throttled run never looks like "this target has no data".

Tips

• Engagement benchmarking: run hashtag or music to discover video ids, then feed those ids into video mode for the full counter set.
• Creator lists: one video call per creator's recent post gives you their follower count, total likes and post count in the same row.
• Bulk url validation: oembed mode is 1.3 KB per target against 302 KB for the embed page — use it to check which of thousands of urls are still live before spending on the heavier mode.
• Sound trend tracking: entityVideoCount on a sound is the total number of videos using it. Re-run on a schedule and diff that number.

Local development

pip install -r requirements.txt
python test_local.py            # all five modes against live TikTok
python test_local.py video      # one mode

The test asserts that required fields per mode are non-null — a mode that returns a row full of nulls fails rather than passing on row count alone.