For AI

Answers

Can ChatGPT or Claude summarize a Bilibili video? (2026)

Short answer

Not by watching it. A chatbot summarizes text, and the text it can get for a Bilibili video through okdata is the title, description, stats and comments: video detail is $0.002 and a comment page $0.002. There is no Bilibili subtitle endpoint, so for what is said in the video you need Bilibili's own AI summary, subtitles fetched with a login, or a transcript you make yourself.

Paste a Bilibili link into ChatGPT or Claude and ask for a summary, and one of two things happens: it says it cannot open the page, or it summarizes whatever text it managed to read, which is rarely the speech in the video. Language models summarize text. The question is which text about a video they can get.

What an AI can read through okdata

Connect the MCP server or call the API from your own code, and the model can read:

Source What the model gets Cost
Bilibili video detail Title, description (text), publish time, uploader, views, likes, favorites, shares, comments count; with view=full also coins, danmaku count, duration and section $0.002
Bilibili video comments About 20 comments a page, hottest or newest first, with likes, reply counts and IP location $0.002 a page

That supports a useful summary of the video's packaging and reception: what the uploader says it is about, how it performed, and what viewers praise, dispute or ask. Some descriptions carry chapter lists, and top comments sometimes restate the key points. What it does not support is a faithful summary of what is said and shown. The catalog has no Bilibili subtitle or transcript endpoint, and danmaku come only as a count.

Where the spoken content can come from

  • Bilibili's own AI summary. The web player shows an AI 视频总结 button under some videos, with a timestamped outline. A write-up on 少数派 notes it does not work on every video.
  • Subtitles. Some videos carry CC or AI subtitles, but yt-dlp's Bilibili extractor warns that some are only available when logged in, in which case a tool needs your account cookie. Summary tools such as BibiGPT have had to chase Bilibili's subtitle route as it changed. bilibili-API-collect, the reference that documented these routes, closed in January 2026 after a lawyer's letter from Bilibili, and the bilibili-api library posted a similar notice.
  • Your own transcript. yt-dlp supports Bilibili; download the audio of a video you are entitled to process and transcribe it locally with Whisper, then give the text to the model.

Example

Over MCP, a prompt like this works: "Read Bilibili video BV17x411w7KC and its 60 hottest comments. Summarize what the uploader says the video covers, how it performed, and what viewers agree and disagree about. Say clearly that you have not seen the video itself." The agent makes four calls, $0.008.

In your own code:

import os, requests

API = "https://api.okdata.online"
H = {"Authorization": f"Bearer {os.environ['OKDATA_API_KEY']}"}

def call(path, **params):
    body = requests.get(API + path, headers=H, params=params, timeout=60).json()
    if body["code"] != "ok":
        raise RuntimeError(f'{body["code"]}: {body["message"]}')
    return body

bv = "BV17x411w7KC"
v = call("/v1/bilibili/videos/get", id=bv, view="full")["data"]["item"]
comments, cursor = [], None
for _ in range(3):
    page = call("/v1/bilibili/videos/comments/list", id=bv, sort="hot",
                fields="text,likes", **({"cursor": cursor} if cursor else {}))
    comments += page["data"]["items"]
    if not page["page"]["has_more"]:
        break
    cursor = page["page"]["next_cursor"]

context = (f"Title: {v['title']}\nDescription: {v['text']}\n"
           f"Views {v['metrics']['views']}, likes {v['metrics']['likes']}, coins {v['ext']['coins']}\n"
           "Top comments:\n" + "\n".join(f"[{c['likes']}] {c['text']}" for c in comments))
# Send `context` to the model of your choice with your summary instructions.

Limits

  • Tell the model what it has not seen. A summary built from a title and comments sounds just as confident as one built from a transcript.
  • Comments come from the most engaged viewers, not the whole audience.
  • Deleted or private videos return not_found, which is still charged as one call.

For YouTube, where transcripts are available, see letting Claude summarize a YouTube video.

Sources

  1. yt-dlp Bilibili extractor: subtitles only available when logged ingithub.com · read
  2. yt-dlp supported sites (BiliBili)github.com · read
  3. openai/whisper: speech recognition modelgithub.com · read
  4. 少数派: AI Summary and Bilibili's AI 视频总结 buttonsspai.com · read
  5. BibiGPT issue 148: Bilibili subtitle API changedgithub.com · read
  6. bilibili-API-collect: repository closed in January 2026github.com · read
  7. Nemo2011/bilibili-api: maintenance stopped, notice from Bilibili's lawyersgithub.com · read

Where people asked this (read 2026-11-12)

Updated

Connect once, then just ask.

Every new key comes with $0.10 to try it. One sentence connects your agent.

Start free
Connect your AI

Paste this into your agent

Read https://okdata.online/agents.md and set up okdata for me; my key is in the OKDATA_API_KEY environment variable.

Claude Code, Cursor, Codex and other agents that can open a link set themselves up from it.

MCP serverhttps://api.okdata.online/mcp