xiaohongshu-mcp Lets AI Post to China's Instagram Without an Official API
A Go-based MCP server wraps Xiaohongshu with browser automation, letting Claude, Cursor, and other AI clients post, search, and engage on the platform.
Takeaways
- Go-based MCP server automating Xiaohongshu via headless browser, wrapping the platform's UI in 13 tools
- Repo has 13.9k stars and 2.1k forks, signaling strong creator demand
- Supports image and video publishing, scheduled posts, feed search with filters, comments, likes, favorites
- Runs at localhost:18060/mcp, integrates with Claude Code, Cursor, VSCode, Cline, Gemini CLI
- Ships prebuilt binaries for macOS/Windows/Linux plus Docker image on Docker Hub
- Sensitive to UI drift and single-session limits; account risk still applies to any automation
Xiaohongshu, the Chinese lifestyle app often described as a blend of Instagram and Pinterest, does not publish an open API for third-party posting. A Go project called xiaohongshu-mcp works around that by wrapping the site in a headless browser and exposing every action through the Model Context Protocol, letting an LLM client drive an account the same way a human would.
The repo has crossed 13.9k stars and 2.1k forks, a strong signal that solo creators and marketing teams have been waiting for a plug-and-play automation layer on top of Xiaohongshu. It also ships with a warning from the maintainer: this is a learning project, and any automation on the platform carries account risk.
13 tools, one browser, zero official API
The MCP server exposes 13 tools once connected. From the documentation, the surface area covers the full life cycle of running a Xiaohongshu account:
check_login_status,get_login_qrcode, anddelete_cookiesfor session managementpublish_contentfor image posts andpublish_with_videofor video posts, both supporting tags, scheduled publishing up to 14 days out, visibility controls, and product bindings for accounts with commerce enabledlist_feedsandsearch_feeds