bilibili-data-mcp 0.2.0

An MCP server that lets any AI client read public Bilibili data: video info, subtitles, danmaku, danmaku hotwords, comments, and search.

bilibili-mcp

An MCP server that lets any AI client (Claude, Cursor, ...) read public Bilibili data: video info, subtitles, danmaku, and search. It's a thin wrapper over the mature bilibili-api library — all WBI signing, cookies, and anti-abuse handling are delegated to it.

Reads public data only. Video info and search work as a guest (no login). Subtitles are gated behind login by Bilibili itself, so the subtitle tool needs your own SESSDATA cookie (see below). For personal/research use at a reasonable request rate.

Tools

Tool	Description
get_video_info(bvid)	Title, uploader, play/like/coin counts, duration, publish date, description. Accepts a BV id or a full video URL. Guest, no login.
get_video_subtitle(bvid, lang="zh-CN", with_timestamp=False)	Full subtitle/transcript text — great for "summarize this video". Requires login (BILI_SESSDATA, see below); returns clear guidance if no cookie is set, and a clear message when a video simply has no subtitles.
get_video_danmaku(bvid, limit=200, with_timestamp=False)	All danmaku (弹幕, on-screen bullet comments), ordered by appearance time — great for asking the AI what memes viewers spam, the overall mood, or which moments are "高能". Guest, no login.
get_danmaku_hotwords(bvid, top=20, segments=5)	Aggregates danmaku into a top repeated-comment ranking plus the highest-density "高能" time segments — a quick read on what viewers spam and which moments spike. Guest, no login.
get_video_comments(bvid, sort="hot", limit=20)	Top-level video comments (username, likes, content), sorted by likes (hot) or newest (time) — great for summarizing reception or finding debate. Guest, no login.
search_videos(keyword, page=1, limit=10)	Search videos by keyword. Guest, no login.

Install

The PyPI package is bilibili-data-mcp (the import module is bilibili_mcp).

Option A — uvx (recommended, no clone, no manual venv)

If you have uv, just point your MCP client at the package — uvx downloads it from PyPI, builds an isolated environment, and runs it automatically:

{
  "mcpServers": {
    "bilibili": {
      "command": "uvx",
      "args": ["bilibili-data-mcp"]
    }
  }
}

To pin a version, use "args": ["bilibili-data-mcp@0.2.0"]. The same config works for any MCP client (Claude Desktop, Cursor, ...) via its claude_desktop_config.json / equivalent.

Option B — from source (development)

git clone https://github.com/haotongliu58-sudo/bilibili-mcp
cd bilibili-mcp
python -m venv .venv
.venv\Scripts\python.exe -m pip install -e .

Then point your MCP client at the built executable:

{
  "mcpServers": {
    "bilibili": {
      "command": "C:\\path\\to\\bilibili-mcp\\.venv\\Scripts\\bilibili-data-mcp.exe"
    }
  }
}

Subtitles & login (BILI_SESSDATA)

Bilibili restricts subtitle data to logged-in users, so get_video_subtitle needs your own SESSDATA cookie. Without it, info and search still work; the subtitle tool just returns a short note telling you to set the cookie.

To enable subtitles, get the SESSDATA value from your browser cookies on bilibili.com (DevTools → Application → Cookies) and pass it as an environment variable to the server:

{
  "mcpServers": {
    "bilibili": {
      "command": "uvx",
      "args": ["bilibili-data-mcp"],
      "env": { "BILI_SESSDATA": "your_sessdata_value_here" }
    }
  }
}

The cookie is read locally and used only to call Bilibili's API directly from your machine — it is never uploaded anywhere. Optional companions (rarely needed for read-only subtitle access): BILI_BILI_JCT, BILI_BUVID3, BILI_DEDEUSERID.

Proxy

By default the server ignores the system proxy and connects directly (Bilibili is a domestic CN site; routing through an overseas proxy node breaks requests). If you are outside mainland China and need a proxy to reach Bilibili, set BILI_USE_PROXY=1 in the server's environment.

License

MIT