Chinese social media · MCP server
Let Claude Code read Xiaohongshu and Zhihu
Web2MD connects MCP agents to your real Chrome session and converts Xiaohongshu, Zhihu, WeChat, and Bilibili pages into structured Markdown. Use it for focused, authorized research when a server-side fetch returns an app prompt, a login shell, or incomplete content.
The short answer
Install the Web2MD MCP server and Agent Bridge, then give your agent a list of URLs. The agent calls the browser extension, which reads each rendered page and returns Markdown with the useful structure preserved. No Xiaohongshu, Zhihu, or WeChat API key is required because the conversion happens in the browser you already use.
Supported Chinese research sources
| Platform | Page types | Markdown output |
|---|---|---|
| Xiaohongshu | Notes and discovery pages | Title, author, note text, engagement metadata, images, and source URL when available |
| Zhihu | Questions, answers, and column articles | Question or article structure, answer text, code blocks, formulas, images, and links |
| Public-account articles | Article title, account, publication context, body, images, and source URL | |
| Bilibili | Video pages with available page metadata or captions | Video metadata and available textual content in Markdown |
Single page
agent_convert({
"url": "https://www.zhihu.com/question/..."
})Best when the agent has found one discussion, note, or article that needs full-text analysis.
Research batch
agent_batch_convert({
"urls": ["https://...", "https://..."]
})Best for a curated collection of up to 50 URLs that the agent will cluster, compare, summarize, or add to a knowledge base.
A practical Claude Code workflow
- Open Chrome and sign in only to the platforms you are authorized to use.
- Run the Web2MD MCP setup and restart Chrome so the native host is registered.
- Give Claude Code or Cursor a focused URL list and ask it to call agent_batch_convert.
- Ask the agent to cite each source URL when it compares themes or extracts claims.
- Keep the Markdown as a reproducible research corpus instead of relying on an untraceable summary.
Responsible use
Agent Bridge is a reader for pages available in your own browser session, not an access bypass. Respect platform terms, copyright, personal information, rate limits, and the expectations of private communities. For unattended or high-volume public crawling, use a crawler designed for that architecture instead.
Chinese social media MCP FAQ
Can Claude Code read Xiaohongshu and Zhihu pages through MCP?
Yes. Web2MD gives MCP clients an agent_convert tool for one URL and agent_batch_convert for up to 50 URLs. The tools route conversion through the Web2MD extension in your real Chrome session and return Markdown to the agent.
Why use a browser session for Chinese social media research?
Many Chinese platforms render important content in the browser, show app-open overlays, or return different content to a datacenter fetch. Reading the page already rendered in your browser gives the extractor the same authorized view that you can see.
Which Chinese platforms have dedicated extractors?
Web2MD includes dedicated handling for Xiaohongshu, Zhihu, WeChat public articles, and Bilibili, along with general browser extraction for other pages. Output quality still depends on the page type and the access available in your session.
Does this bypass login, anti-bot systems, or private access?
No. It reads content that is already available in your browser session. It does not defeat authentication, solve access challenges automatically, or grant permission to content you cannot view. You remain responsible for the platform terms and how you use the output.
Can I batch-convert Xiaohongshu or Zhihu URLs?
Yes, up to 50 URLs per agent_batch_convert call. They are processed sequentially through Chrome, so this is designed for focused research collections rather than unattended large-scale crawling.