Automating China's Instagram Without an Official API

What it does Wraps the Chinese lifestyle platform Xiaohongshu in a Model Context Protocol server, exposing login, publishing, search, and social interaction as tools for AI assistants. It handles text-and-image posts, video uploads, comments, likes, favorites, and profile scraping—essentially the full surface area of a human user.

The interesting bit Xiaohongshu offers no official API, so the project drives a headless browser to simulate real user behavior. The author claims a predecessor automation ran for over a year without bans, and this MCP version was validated with Claude Code for weeks before release.

Key highlights

• Full read-write access: publish posts (images or local video), search feeds, read comments, reply, like, and favorite
• Anti-duplicate smarts: like and favorite tools check current state before acting to avoid redundant requests
• Browser plugin alternative (x-mcp) available for users who want zero configuration
• All donations go to charity; the project is explicit about its educational purpose
• Account restrictions apply: new accounts may need real-name verification, and simultaneous web logins kick the MCP session offline

Caveats

• First run downloads a ~150MB headless browser
• Many actions require both a post ID and an xsec_token scraped from feed or search results
• Video uploads only work from local file paths, not URLs
• The README notes a daily posting ceiling of roughly 50 posts and warns that content must avoid banned words and reposting

Verdict Social media managers and developers building AI agents for Chinese platforms should look here; anyone needing a stable, officially supported API should look elsewhere.