Bilibili 哔哩哔哩 Video Scraper

The Bilibili 哔哩哔哩 Video Scraper extracts public data from Bilibili, China's largest video community. One actor, nine modes: search videos, read a video's full stats, a creator's profile, uploads, followers and feed, comments with replies, on-screen danmaku, and the video or audio file itself.

中文说明

哔哩哔哩（Bilibili）公开数据抓取工具。按关键词搜索视频，或用 BV 号、UP 主 mid 获取视频详情、UP 主资料与投稿、评论与回复、粉丝、动态、弹幕，也可下载视频或音频文件。

每行数据都带 kind 标明类型：视频含标题 title、UP 主 owner_name、播放 view、点赞 like、评论数 reply；评论含正文 message 与作者 member_name；弹幕含正文 message 与出现时间 progress_ms；下载文件给出 download_url。

运行自带住宅代理，无需自行配置，多数模式也无需登录。在 Apify Console 选好模式、填入关键词或 ID 后点击 Start 即可，也可以通过 Apify API 提交同样的 JSON。结果支持导出 JSON、CSV 或 Excel。

What you get

One dataset, one row per record, with kind saying what each row is. Export as JSON, CSV or Excel.

{
  "kind": "video",
  "bvid": "BV1wD4y1o7AS",
  "title": "花了2万多买的Python教程全套，现在分享给大家，入门到精通(Python全栈开发教程)",
  "owner_name": "Python_子木",
  "view": 35494331,
  "like": 1094798,
  "reply": 23760,
  "scraped_at": "2026-09-04T11:20:31.004112+00:00"
}

Every mode writes into the same dataset, so a mixed run filters cleanly on kind:

• video — bvid, title, owner_name, view, like, coin, favorite, reply, danmaku, duration_sec, published_at, tags, cover_url
• creator — mid, name, sign, follower, following
• comment — rpid, parent_rpid, member_name, message, like, reply_count, ctime
• fan — creator_mid, mid, name, followed_at
• dynamic — dynamic_id, dynamic_type, text, bvid, like, comment, forward
• danmaku — cid, part, parts_total, bvid, progress_ms, message, mode, color, sender_hash
• media — bvid, media, quality_id, codec, size_bytes, download_url

A video row also says which target found it: source_keyword for a search, source_creator_mid for a creator's uploads. Pass several keywords or creators in one run and split the rows on those columns.

download_url points at the file in this run's storage, so download anything you mean to keep before the run's data retention expires.

What it costs

Event	Price
Actor Start (apify-actor-start)	$0.00005
video (video)	$0.004
creator (creator)	$0.004
comment (comment)	$0.001
fan (fan)	$0.001
dynamic (dynamic)	$0.0015
danmaku (danmaku)	$0.00002
danmaku, per 1,000 (danmaku-1k)	$0.01
media (media)	$0.002
media, per 10 MB (media-file-10mb)	$0.002

Input

Field	Type	Required	Default	What it does
mode	string	no	"search"	Which kind of record to collect: search finds videos by keyword; video returns full stats for specific videos; creator returns a creator's profile and follower count; creator_videos lists a creator's uploads; comments returns a video's comments; fans lists a creator's followers; dynamics reads a creator's feed; danmaku returns a video's on-screen bullet comments; media returns a download link for the video or its audio. Default search.
keywords	array	no	—	Search terms, one per entry — for example python. Each term returns up to max_items videos, ranked the way order says. Bilibili search ignores capitals, so terms that differ only in capitals or surrounding spaces are searched, and billed, once. Used when mode is search; leave empty for every other mode.
video_ids	array	no	—	Videos to work on, one per entry, as a BV id (BV1wD4y1o7AS), an av id (av712020469) or the video part of a bilibili.com URL — the two id styles name the same video and both work. Used when mode is video, comments, danmaku or media; leave empty for every other mode.
creator_ids	array	no	—	Creators to work on, one numeric mid per entry — the number in a space.bilibili.com/12345 URL. Used when mode is creator, creator_videos, fans or dynamics; leave empty for every other mode.
max_items	integer	no	100	How many records to return per keyword, video or creator, at least 1. Default 100. Comments: counts top-level comments only; with include_replies their replies (up to max_replies_per_comment each) come on top, so a run can return and bill more rows than this. Danmaku: applies to each six-minute stretch of a video, not the whole video, and with danmaku_all_parts to each stretch of every part read, so a long video returns more rows than this.
order	string	no	"totalrank"	How search results are ranked: totalrank for best match, click for most viewed, pubdate for newest, dm for most bullet comments, stow for most favourited. Default totalrank. Used when mode is search.
duration	string	no	"any"	Keep only search results of this length: any for no filter, under_10m for under ten minutes, 10_30m, 30_60m, over_60m for over an hour. Default any. Used when mode is search.
tid	integer	no	—	Bilibili category to restrict a search to, given as the numeric tid that appears in a category page's URL. Leave empty to search every category. Used when mode is search.
include_replies	boolean	no	false	Also collect the replies under each comment, as extra rows whose parent_rpid names the comment they belong to. Default false, which returns top-level comments only. Used when mode is comments.
max_replies_per_comment	integer	no	20	Most replies to collect under one comment, at least 1. Default 20. It does nothing unless include_replies is true.
danmaku_all_parts	boolean	no	false	Read danmaku from every part of a multi-part video (a course or series uploaded as one video), up to max_parts parts, instead of part 1 only. Each row's part and parts_total say which part it came from out of how many (part 3 of 200). Default false. Used when mode is danmaku.
max_parts	integer	no	10	Most parts of one multi-part video to read danmaku from, counting from part 1, at least 1. Default 10. Every part adds its own requests and rows, so raise it with care on long courses. It does nothing unless danmaku_all_parts is true.
quality	string	no	"best"	Which stream to download: best for the highest the video offers, 720p or 480p to cap it at that resolution. Default best. A video that offers nothing at the resolution you ask for returns the best one below it, never a larger file than you asked for. Used when mode is media.
audio_only	boolean	no	false	Download the audio track on its own instead of the video file — far smaller, and enough for transcription or music. Default false. Used when mode is media.
auth	object	no	—	Your own Bilibili cookies, as {"cookies": "SESSDATA=..."}, if you want the run to act as your own account — for instance to reach media resolutions Bilibili reserves for members. Leave empty and you need no account at all.
proxyUrls	array	no	—	Leave empty and the run uses the residential proxy this actor ships with, included in the price of the run. To route the run through your own account instead, add one gateway URL per entry, for example http://user:pass@host:port — works with DataImpulse, Bright Data, Oxylabs, Smartproxy or any provider that issues URLs. When set, only these URLs are used.

How to run it

1. Open the actor on Apify Console and pick a What to scrape mode — it decides which other fields matter.
2. Fill that mode's target: keywords for search, video IDs for video, comments, danmaku and media, creator IDs for creator, creator videos, fans and dynamics.
3. Run. Export JSON, CSV or Excel from the dataset.

{
  "mode": "search",
  "keywords": ["python", "机器学习"],
  "order": "click",
  "duration": "10_30m",
  "max_items": 200
}

Minimal input:

{
  "keywords": ["python"]
}

Available as the MCP tool herus13--bilibili-scraper on mcp.apify.com and through the Apify API; send the same JSON.

Why this scraper

• Nine modes, one actor and one dataset. Search, video stats, creator profile, uploads, comments, followers, feed, danmaku and media downloads all write rows tagged with kind, so a whole research pass is one run rather than five actors and a join.
• Danmaku, which is the Bilibili-specific signal. The on-screen comment track comes with progress_ms, pinning each remark to the second of the video it reacts to — audience reaction with timing attached, which no comment feed gives you.
• No account needed. Seven of the nine modes run anonymously, and you can still paste your own cookies if you want a run to act as you.
• BV or av, either works. Paste whichever id a page gave you; av712020469 and BV1wD4y1o7AS are the same video.
• Media as a playable link. The video or the audio track alone, hosted as a link that works outside Bilibili.
• Nothing to set up. A residential proxy is included in the price of the run.

Use cases

• Content and trend research — What ranks for a keyword today, ordered by views or recency, and how a video's stats move when you re-run the search on a schedule.
• Creator benchmarking — A creator's whole upload history with per-video engagement, next to their follower count, for outreach shortlists or competitor tracking.
• Audience sentiment — Comments and their replies for considered opinion, danmaku for the unfiltered live reaction, both as plain text ready for a classifier.
• Reaction timing analysis — Group danmaku by progress_ms to find the exact moments of a video that made an audience respond.
• Media archiving and transcription — Pull the audio track alone and feed it to a transcription step, or keep the video file for an archive.

FAQ

Do I need a Bilibili account? No. Seven of the nine modes run anonymously. If you would rather run as yourself — for the media resolutions Bilibili reserves for members, for instance — paste your own cookies into Your own login cookies.

Can I pass a video URL instead of an id? Use the BV id from the URL: https://www.bilibili.com/video/BV1wD4y1o7AS means BV1wD4y1o7AS. Old-style av ids such as av712020469 work too.

Why did I get fewer comments than the video reports? Bilibili publishes a sample of a popular video's comments to any single visitor, not the whole archive. A video showing tens of thousands of comments will not return tens of thousands of rows.

Why did a long video return more danmaku than Max items? Bilibili serves danmaku in six-minute stretches, and the cap applies to each stretch rather than to the whole video. With Danmaku from every part on, it applies to each stretch of every part read.

Why did a comments run return more rows than Max items? Max items counts top-level comments. With Include replies on, each comment's replies (up to Max replies per comment) come on top as their own rows, and every row is billed as a comment.

Why did a multi-part video return danmaku from part 1 only? A course or series uploaded as one video has one danmaku track per part. Part 1 is read by default; turn on Danmaku from every part and set Max parts per video to read more, and each row's part and parts_total say which part it came from out of how many. When Max parts per video leaves parts unread, the run log says how many were skipped.

Is a proxy included? A residential proxy is included in the price of the run. To use your own, paste gateway URLs into Your own proxy URLs. Scraping at volume? Your own DataImpulse account is cheaper per GB.

Is scraping Bilibili legal? This actor reads only public pages, the ones any visitor sees, and never signs in as you unless you supply cookies yourself. You are responsible for how you use the data, including copyright in downloaded media and personal data in comments. Check Bilibili's terms and your local rules.

Related actors

Building a wider video or social pipeline? Pair this actor with:

• TikTok Video Scraper — video metadata, stats and optional video files.
• TikTok Profile Scraper — creator profiles with follower and engagement counts.
• YouTube Transcript Scraper — transcripts with timestamps, for the same research in another market.
• Instagram Post Scraper — posts with captions, hashtags and engagement.