Likes, favourites and comments come back with every note, so a search can be sorted by what an audience actually responded to instead of by what the platform chose to show you. It works by driving a real browser, which is why the first call is slow and why concurrency is capped at 3 — both are deliberate, not faults. The repository frames this as for research and education and asks you to respect the platform's terms, which is worth taking at face value before pointing it at anything at scale.
A TypeScript server that drives a Playwright browser against Xiaohongshu, running keyword searches and extracting each note into structured fields.
- Keyword search across Xiaohongshu, with results returned as data rather than a page of HTML
- Per note: title, content, author name and description, and the direct link
- Interaction metrics — likes, favourites and comment counts — so you can tell what actually landed
- Hashtags and image URLs, in WebP format
- Login state saved to a cookie store and reused, so you sign in once
- Concurrent retrieval, capped at 3 parallel requests to stay under the platform's limits
A Xiaohongshu account — the first run opens a browser for you to log in manually, and the cookies are reused afterwards. Node.js 18 or higher with pnpm, and Playwright browsers installed. Expect the first search of a session to take 30 to 60 seconds while the browser starts and the page loads, so raise your client's timeout before assuming it hung.
One command — npx -y rednote-mcp