XHS AI Toolkit

Make AI understand your Xiaohongshu (RedNote)

简体中文 | English

AI-powered toolkit for Xiaohongshu (小红书 / RedNote) that turns your favorite posts into AI memory.

• MCP Integration — Search, browse, comment via AI assistants
• Trend Tracking — Auto-generate topic reports with engagement analytics
• Memory Export — Convert your liked/saved posts into AI-searchable knowledge base

Built on xiaohongshu-mcp and XHS-Downloader.

Features

Quick Start

1. Install xiaohongshu-mcp

Download from GitHub Releases:

# Linux x64
wget https://github.com/xpzouying/xiaohongshu-mcp/releases/latest/download/xiaohongshu-mcp-linux-amd64.tar.gz
wget https://github.com/xpzouying/xiaohongshu-mcp/releases/latest/download/xiaohongshu-login-linux-amd64.tar.gz

# macOS ARM
wget https://github.com/xpzouying/xiaohongshu-mcp/releases/latest/download/xiaohongshu-mcp-darwin-arm64.tar.gz
wget https://github.com/xpzouying/xiaohongshu-mcp/releases/latest/download/xiaohongshu-login-darwin-arm64.tar.gz

Install:

mkdir -p ~/.local/bin
tar -xzf xiaohongshu-mcp-*.tar.gz -C ~/.local/bin/
tar -xzf xiaohongshu-login-*.tar.gz -C ~/.local/bin/

cd ~/.local/bin
mv xiaohongshu-mcp-* xiaohongshu-mcp
mv xiaohongshu-login-* xiaohongshu-login
chmod +x xiaohongshu-mcp xiaohongshu-login

2. Install This Toolkit

# Clone to OpenClaw workspace
git clone https://github.com/zhjiang22/openclaw-xhs.git
cp -r openclaw-xhs ~/.openclaw/workspace/skills/xiaohongshu

# Or use symlink
ln -s /path/to/openclaw-xhs ~/.openclaw/workspace/skills/xiaohongshu

# Verify installation
cd ~/.openclaw/workspace/skills/xiaohongshu/scripts
./install-check.sh

3. Login (Get Cookies)

Option A: Desktop Environment

./login.sh  # Opens browser, scan QR code with Xiaohongshu app

Option B: Headless Server

Get cookies on your local machine, then copy to server:

# On local machine with GUI
./xiaohongshu-login
# Cookies saved to /tmp/cookies.json

# Copy to server
scp /tmp/cookies.json user@server:~/.xiaohongshu/cookies.json

4. Start Service

./start-mcp.sh              # Headless mode
./start-mcp.sh --headless=false  # Show browser (debug)

Service runs at http://localhost:18060/mcp.

Server Deployment (Headless Linux)

On servers without a desktop environment, the underlying browser requires a virtual display. start-mcp.sh auto-detects the environment — if no display is found, it starts Xvfb automatically. Just install it first:

# Debian/Ubuntu
sudo apt-get install -y xvfb

# CentOS/RHEL
sudo yum install -y xorg-x11-server-Xvfb

No extra configuration needed. The script handles:

• Detecting the DISPLAY environment variable
• Auto-starting Xvfb :99 when no display is available
• Cleaning up Xvfb when stop-mcp.sh is called

Note: Without Xvfb, login and search will fail on headless servers. See Issue #3.

Usage

Basic Commands

./status.sh                    # Check login status
./search.sh "coffee"           # Search posts
./recommend.sh                 # Get recommendations
./post-detail.sh <id> <token>  # Get post details
./comment.sh <id> <token> "Great post!"  # Comment
./user-profile.sh <user_id> <xsec_token>  # Get user profile

Trend Tracking

Auto-search trending posts and generate analysis reports:

./track-topic.sh "AI" --limit 10
./track-topic.sh "travel" --limit 5 --output report.md
./track-topic.sh "iPhone" --limit 5 --feishu  # Export to Feishu

MCP Tools

Long Image Export

Export posts as annotated JPG long images (white background, black text):

# Prepare posts.json
cat > posts.json << 'EOF'
[
  {
    "title": "Post title",
    "author": "Author",
    "stats": "13k likes 100 saves",
    "desc": "Post summary",
    "images": ["https://...webp"],
    "per_image_text": {"1": "Caption for 2nd image"}
  }
]
EOF

./export-long-image.sh --posts-file posts.json -o output.jpg

Requires: Python 3.10+, Pillow (pip install Pillow)

Memory Export (Turn Likes into AI Memory)

Export your liked/saved posts as a searchable knowledge base for AI assistants.

1. Install XHS-Downloader

git clone https://github.com/JoeanAmier/XHS-Downloader.git
cd XHS-Downloader
pip install -r requirements.txt

2. Extract Post Links (Tampermonkey Script)

1. Install Tampermonkey
2. Install XHS-Downloader UserScript
3. Go to Xiaohongshu web → Profile → Liked/Saved
4. Click Tampermonkey menu → "Extract liked posts" or "Extract saved posts"
5. Links auto-copied to clipboard
6. Paste into links.md

3. Download & Export

# Copy helper scripts
cp tools/xhs-downloader/*.py /path/to/XHS-Downloader/

# Download posts
cd /path/to/XHS-Downloader
python batch_download.py links.md

# Export to workspace
python export_to_workspace.py
# Output: ~/.openclaw/workspace/xhs-memory/

4. Configure OpenClaw Memory Search

Edit ~/.openclaw/openclaw.json:

{
  "memorySearch": {
    "extraPaths": [
      "~/.openclaw/workspace/xhs-memory"
    ]
  }
}

Now your AI assistant can search your Xiaohongshu favorites!

Project Structure

openclaw-xhs/
├── README.md             # English docs
├── README_CN.md          # Chinese docs
├── LICENSE
├── SKILL.md              # Skill manifest
├── scripts/              # MCP wrapper scripts
│   ├── install-check.sh
│   ├── start-mcp.sh
│   ├── stop-mcp.sh
│   ├── login.sh
│   ├── mcp-call.sh
│   ├── status.sh
│   ├── search.sh
│   ├── recommend.sh
│   ├── post-detail.sh
│   ├── comment.sh
│   ├── user-profile.sh
│   ├── track-topic.sh
│   ├── track-topic.py
│   ├── export-long-image.sh
│   └── export-long-image.py
└── tools/
    └── xhs-downloader/   # Memory export tools
        ├── README.md
        ├── batch_download.py
        ├── export_memory.py
        └── export_to_workspace.py

Security

This project implements the following security measures:

• Cookie protection: Cookie files are copied with 600 permissions (owner-only read/write)
• Injection prevention: All shell scripts use jq to build JSON payloads instead of string interpolation, preventing shell injection
• Tool name validation: MCP tool names are restricted to alphanumeric characters and underscores
• Path validation: Cross-skill script calls validate that target paths are within allowed directories
• Third-party content: Content fetched from Xiaohongshu is user-generated; exercise appropriate caution

Disclaimer

This project is a wrapper layer for xiaohongshu-mcp.

• Does NOT contain xiaohongshu-mcp source code
• Users must download xiaohongshu-mcp binaries separately
• Scripts communicate via HTTP protocol only

Acknowledgments

• @xpzouying — xiaohongshu-mcp
• @JoeanAmier — XHS-Downloader (GPL-3.0)

License

MIT License (wrapper scripts only)

Note: xiaohongshu-mcp has no declared license. Please respect the author's terms.

If this project helps you, please give it a ⭐!