Xhs Mcp Service

小红书（XHS/RED）MCP 服务。通过本地 xhs-mcp-server 服务提供完整的小红书操作能力。 当用户提到小红书、红书、XHS、RED、发笔记、搜笔记、小红书运营等任何与小红书相关的操作时使用此技能。 ⚠️ 前置条件：需要先启动 xhs-mcp-server 服务（默认 http://localho...

Security Scan

The package and SKILL.md are coherent with a local MCP-based Xiaohongshu automation service, but there are a few mismatches and exposure risks (notably a server bind default of 0.0.0.0 vs the SKILL.md's localhost claim and handling of session cookies) that you should review before installing or running.

Purpose & Capability

Name/description claim a local xhs-mcp-server to operate Xiaohongshu; the repository contains code (Express + MCP SDK + Puppeteer) implementing exactly that functionality (13 tools: search, like, publish, etc.). The requested files, dependencies (puppeteer, express, @modelcontextprotocol/sdk), and runtime behaviors are proportionate to the stated purpose.

Instruction Scope

SKILL.md instructs installing deps, running login (opens a browser for QR code), starting the local MCP service, and running scripts that read local files and publish content. Those instructions are expected, but notable issues: SKILL.md claims the service endpoint is http://localhost:18060/mcp, yet the server code defaults XHS_HOST to '0.0.0.0' (bind all interfaces) — this mismatches the documentation and could expose the MCP endpoints beyond localhost. The instructions also tell users to run commands in a hard-coded Windows path (D:\work\...), which is a fragile/poorly documented instruction (example path) but not necessarily malicious.

Install Mechanism

This is an instruction-only skill for the agent, with included source files (no automated installer). There is no opaque download URL; package.json declares standard npm dependencies (puppeteer, express, MCP SDK). Puppeteer will download Chromium during npm install (normal behavior). No high-risk remote install steps are embedded in SKILL.md or package.json.

Credentials

The skill declares no required environment variables or credentials. It does use/mention XHS_PROXY and XHS_PORT/XHS_HOST environment variables to configure runtime behavior — appropriate for a local service. The code reads/writes data/cookies.json to persist session cookies; this is necessary for automating a user account but means sensitive authentication cookies are stored on disk. That is expected for the purpose, but it is sensitive and should be treated accordingly.

Persistence & Privilege

The skill is not declared always:true. However, a real risk exists: server code defaults XHS_HOST to '0.0.0.0' (listening on all interfaces) which can expose the MCP API (and thereby operations on the user's Xiaohongshu account via stored cookies) to other machines on the network if firewall/host config is not tightened. This network exposure is inconsistent with the SKILL.md's repeated reference to http://localhost and raises an operational security concern.

What to consider before installing

What to consider before installing/running: - This package implements a local automation server that will control your Xiaohongshu account via Puppeteer and stored cookies. If you run it, npm install will download Chromium (Puppeteer) and the service will persist session cookies to data/cookies.json — treat that file as highly sensitive (it contains authentication cookies). Keep it on a trusted machine and with tight file permissions. - The documentation repeatedly refers to localhost, but the server code defaults to binding XHS_HOST='0.0.0.0' which will accept connections from any network interface. Unless you intentionally need remote access, set XHS_HOST=127.0.0.1 (or the equivalent) before starting, or firewall the port. Exposing the MCP endpoint on the network could allow others to control your account if they can reach the host. - The code performs real actions (likes, comments, publishes) against the service using your account. Use test accounts for experimentation and be cautious about automated publishing or bulk operations (risk of platform restrictions/ban). - Review the source (especially src/xhs-tools.js and any code that performs network requests) yourself to ensure there are no unexpected external callbacks or telemetry endpoints. Confirm there are no hardcoded remote endpoints that exfiltrate cookies or data. - Because SKILL.md uses a hard-coded example Windows path and recommends running login which opens a browser for QR scan, run the project in a controlled environment (VM, container, or isolated host) if you are unsure. - If you decide to run: (1) audit code, (2) set XHS_HOST=127.0.0.1, (3) restrict port via firewall, (4) protect data/cookies.json, and (5) prefer using a disposable/test account for initial trials. If you want, I can point to the exact lines/files that set the host binding and the cookies path so you (or your admin) can change them before running.

你是小红书操作助手，帮助用户通过 MCP 协议操作小红书。

🔧 部署要求

必须先完成以下步骤：

1. 进入项目目录

   cd D:\work\xiaohongshu\xiaohongshu-mcp-node
   

2. 安装依赖

   npm install
   

3. 首次登录小红书

   npm run login
   

   在弹出的浏览器中扫码登录，登录成功后 cookies 会自动保存。

4. 启动 MCP 服务

   npm start
   

   服务将运行在 http://localhost:18060/mcp

📋 功能列表（13个工具）

工具	说明	参数
check_login_status	检查登录状态	无
get_login_qrcode	获取登录二维码	无
delete_cookies	删除登录状态	无
list_feeds	获取首页推荐	无
search_feeds	搜索笔记	keyword, filters?
get_feed_detail	获取笔记详情	feed_id, xsec_token
like_feed	点赞/取消点赞	feed_id, xsec_token, unlike?
favorite_feed	收藏/取消收藏	feed_id, xsec_token, unfavorite?
post_comment_to_feed	发表评论	feed_id, xsec_token, content
reply_comment_in_feed	回复评论	feed_id, xsec_token, content, comment_id?, user_id?
user_profile	获取用户主页	user_id, xsec_token
publish_content	发布图文	title, content, images, tags?
publish_with_video	发布视频	title, content, video, tags?

🔧 通过 MCPorter 调用

如果 OpenClaw 已配置 MCPorter，可以这样调用：

# 添加 MCP 服务
mcporter config add xhs-mcp http://localhost:18060/mcp

# 检查登录状态
mcporter call xhs-mcp.check_login_status

# 搜索笔记
mcporter call xhs-mcp.search_feeds keyword="美食"

# 获取首页推荐
mcporter call xhs-mcp.list_feeds

# 点赞（需要 feed_id 和 xsec_token）
mcporter call xhs-mcp.like_feed feed_id="xxx" xsec_token="xxx"

⚠️ MCPorter 限制

MCPorter 在 Windows 命令行下无法传递数组参数，因此：

• ✅ 简单参数功能：可用 MCPorter（搜索、点赞、收藏、评论等）
• ❌ 数组参数功能：需要用 Node.js 脚本（发布图文、发布视频）

📝 发布图文/视频

由于 MCPorter 不支持数组参数，发布功能需要用 Node.js 脚本：

// 创建发布脚本 publish.js
import { publishContent } from './src/xhs-tools.js';

await publishContent({
  title: "标题（最多20字）",
  content: "正文内容（最多1000字）",
  images: ["图片1.jpg", "图片2.jpg"],
  tags: ["标签1", "标签2"]
});

运行：

node publish.js

🎯 使用场景

场景1：检查登录状态

用户：帮我检查小红书登录状态
→ mcporter call xhs-mcp.check_login_status

场景2：搜索笔记

用户：搜索小红书上关于咖啡的内容
→ mcporter call xhs-mcp.search_feeds keyword="咖啡"

场景3：点赞收藏

用户：帮我点赞搜索结果的第一篇
→ 1. 搜索获取 feed_id 和 xsec_token
→ 2. mcporter call xhs-mcp.like_feed feed_id=xxx xsec_token=xxx

场景4：发布笔记

用户：帮我发布一篇小红书笔记
→ 需要用 Node.js 脚本（MCPorter 不支持数组参数）
→ node publish.js

📁 项目文件位置

• 项目目录: D:\work\xiaohongshu\xiaohongshu-mcp-node\
• 主程序: src/index.js
• 登录工具: src/login.js
• 工具实现: src/xhs-tools.js
• 浏览器管理: src/browser.js
• Cookies 存储: data/cookies.json

🔧 环境变量

变量	说明	默认值
XHS_PORT	服务端口	18060
XHS_HOST	绑定地址	0.0.0.0
XHS_PROXY	代理地址	-

使用代理启动：

XHS_PROXY=http://proxy:port npm start

⚠️ 重要注意事项

1. 账号安全：同一账号不能在多个网页端登录，会互相踢出
2. 发布限制：每天最多 50 篇笔记
3. 标题限制：最多 20 字
4. 正文限制：最多 1000 字
5. Cookie 有效期：Cookie 可能过期，需要定期重新登录

💡 最佳实践

1. 简单操作用 MCPorter：搜索、点赞、收藏、评论
2. 复杂操作用 Node.js：发布图文、发布视频、批量操作
3. 定时发布：结合 cron 技能定时调用发布脚本
4. 错误处理：如果操作失败，先检查登录状态

🔗 相关资源

• MCP 协议: https://modelcontextprotocol.io/
• MCP Inspector: npx @modelcontextprotocol/inspector