Xiaohongshu MCP Server

The Xiaohongshu MCP service provides both the Model Context Protocol (MCP) and a standard HTTP REST API. Developers can interact with the service using the HTTP API for automation and integration tasks.

Base URL: http://localhost:18060

Depending on your technical expertise and requirements, you can choose between two primary ways to integrate Xiaohongshu data with your AI assistant:

1. Deep Openclaw Integration (Recommended for Developers): Use this if you want to deploy the MCP server yourself. You can use either xiaohongshu-mcp-skills (if you have already deployed this project) or xiaohongshu-skills (for out-of-the-box usage).
2. x-mcp Browser Extension (Recommended for Non-Technical Users): This is the simplest setup. It is a zero-configuration browser extension (Chrome/Edge) that runs on your local network. It avoids Docker, proxy, and environment setup, and eliminates server IP risks.

In addition to standard HTTP APIs, this service supports the Model Context Protocol (MCP). You can interact with the service using MCP clients via the following endpoints:

• MCP Endpoints: /mcp and /mcp/*path
• Protocol Type: Supports JSON response format via Streamable HTTP.

You can publish videos to Xiaohongshu with a title, description, and tags.

Requirements and Constraints:

• Input Type: Only supports local video file absolute paths. HTTP links are not supported.
• File Size: It is recommended to keep video files under 1GB.
• Process: The system automatically handles format conversion and waits for processing to complete before publishing.

Example input format:

"/Users/username/Videos/video.mp4"

When running in Docker, pay attention to these two storage requirements:

1. Image Publishing: The service generates an images/ directory for published images. This directory is mounted into the container. If you want to publish local images, copy them to the ./images/ directory on your host and instruct the MCP to use the path /app/images during publishing.
2. Data Persistence: The Docker image includes a built-in browser and pre-downloads data during the build stage. To persist cookies and runtime data, you must mount the host directory to the container using ./data:/app/data.

You can publish posts containing both text and images. The post must follow Xiaohongshu's constraints:

• Title: Maximum 20 characters.
• Content: Maximum 1000 characters.

Images can be provided in two ways:

1. HTTP/HTTPS links: e.g., ["https://example.com/image1.jpg"].
2. Local absolute paths (Recommended): e.g., ["/Users/username/Pictures/image1.jpg"]. Using local paths is more stable, faster, and supports more formats.

Supported features include adding appropriate Tags to increase traffic.

["/Users/username/Pictures/image1.jpg", "/home/user/images/image2.png"]

The post-to-xhs skill supports two distinct posting modes based on the type of content you want to share:

1. Upload Image/Text (Default): Best for daily sharing. It combines images with short text.
2. Write Long Article: Best for deep, long-form content. It uses a long article format combined with layout templates. Use this mode by explicitly saying "发长文" (Post long article).

Input can be provided as Full Content (Title + Body + Images/URLs) or via URL Extraction (providing a webpage URL which the system will attempt to fetch and summarize).

All API responses follow a unified JSON structure. Depending on the outcome, you will receive either a success or an error object.

### Success Response
```json
{
  "success": true,
  "data": {},
  "message": "操作成功消息"
}

Error Response

{
  "error": "错误消息",
  "code": "ERROR_CODE",
  "details": "详细错误信息"
}

Operational Limits

• Daily Limit: It is recommended to stay under 50 posts per day.
• Content Quality: Avoid banned words and do not perform traffic diversion or pure content scraping/reposting.

Critical Warnings

• Single Session: RedNote does not allow the same account to be logged in on multiple web platforms simultaneously. Logging into another web platform will "kick out" the MCP session.
• Real-name Verification: New accounts or unverified accounts may trigger a real-name verification prompt. This is a platform requirement and not an account ban; completing verification allows normal operation.
• Account Safety: This project is for learning purposes. Use responsibly to avoid platform crackdowns.

To use Kimi-k2 with Claude Code, you must install the kimi-cc project. During the installation process, the script will pause and prompt you to enter your Kimi API Key. The input will be hidden for security. After installation, you must restart your shell environment or reload your environment variables as instructed by the installer logs to ensure the configuration takes effect.

bash -c "$(curl -fsSL https://raw.githubusercontent.com/LLM-Red-Team/kimi-cc/refs/heads/main/install.sh)"

To use the MCP tools, you must first run the service locally. The service provides an HTTP endpoint at http://localhost:18060/mcp.

• Headless mode (default): Runs without a UI.
• Interface mode: Runs with a user interface.

Ensure you have Go installed to run the service from source.

# Start service (default headless mode)
go run .

# Or with interface mode
go run . -headless=false