Summary
The reason to reach for Baidu specifically is coverage of the Chinese web that Western engines index thinly.
It is a search-and-read pair rather than a research agent: you get results, then you fetch the ones worth reading, and the redirect-cleaning step matters because Baidu's raw result links are otherwise unusable downstream. The throttles are the thing to plan around — 30 searches and 20 fetches per minute, queued rather than dropped, so a wide crawl will simply wait.
What it is
A Python server with two tools: search runs a Baidu query and returns formatted results, fetch_content retrieves a page and extracts its readable text.
What you get
- search — Baidu results as titles, URLs and snippets, with max_results defaulting to 10
- fetch_content — a webpage reduced to clean text, with the boilerplate stripped
- Baidu redirect URLs resolved, so the links you get back point at the real pages
- Ads and irrelevant blocks removed before the results reach the model
- Built-in rate limiting with queueing: 30 search requests per minute, 20 content fetches per minute
Requirements
Nothing to authorize — no account, no key. Published to PyPI as baidu-mcp-server and launched with uvx; a Smithery install path for Claude Desktop is documented as well.
Setup effort
One command — uv pip install baidu-mcp-server