🤖 Xiaohongshu (XHS) MCP Agent System

A full-stack AI Agent system for automating social media tasks (specifically Xiaohongshu). It combines a Next.js Frontend, a FastAPI Backend, and a LangGraph/MCP Agent to control a local Chrome browser.

🌟 System Components

1. Frontend (frontend/):

   • A modern "Canva-like" visual editor built with Next.js 16.
   • Features a drag-and-drop canvas and an AI Chat Assistant.
   • Communicates with the backend to "Publish" posts or "Chat" with the agent.

2. Backend (backend/):

   • A FastAPI server that exposes the Agent's capabilities via HTTP.
   • Endpoints: /chat (for advice) and /publish (for automation).
   • Contains all Python logic (agent_server.py, agent_core.py).

3. The Agent (backend/agent_core.py):

   • Powered by GPT-5 (via AiHubMix) and LangGraph.
   • Hybrid Search Architecture: Intelligently routes queries between fast LLM-based search and deep Browser Automation (via MCP).
   • Uses MCP (Model Context Protocol) to control Chrome via mcp-chrome-bridge.
   • Can navigate, click, extract images, and download files.

4. CLI Tool (backend/agent_chrome.py):

   • A standalone command-line version of the agent for testing and "Auto-Pilot" mode.

---

⚠️ Configuration (Crucial!)

Before running anything, you MUST configure the paths for your machine.

👉 READ THE CONFIGURATION GUIDE HERE 👈

Key items to configure:

1. Path to mcp-server-stdio.js in backend/agent_server.py and backend/agent_chrome.py.
2. backend/searcher_api.txt (API Key).
3. backend/auth.json (Cookies).

---

🚀 Quick Start Guide

Step 1: Start the Backend (Agent)

Open a terminal in the root directory (xhs-mcp-server/):

# Activate your Python environment
# conda activate xhs_env
cd backend
python agent_server.py

The server will start on http://127.0.0.1:8000. It runs in Auto-Pilot Mode, meaning it executes tools automatically without pausing for confirmation.

Step 2: Start the Frontend

Open a new terminal in xhs-mcp-server/:

cd frontend
npm run dev

Open http://localhost:3000 in your browser.

Step 3: Use the System

1. Chat: Use the "AI Assistant" panel on the right to ask the agent to do things (e.g., "Search for design trends", "Add a title saying 'Hello'").
2. Canvas: The agent can now modify your canvas directly!
3. Publish: Click the "Publish" button to let the agent automate the posting process to Xiaohongshu.

---

📚 Documentation

• Developer Guide & API Reference - Start Here for Deep Dives! Detailed analysis of file structure, API contracts, and data flow.
• Configuration Guide - Essential setup steps (Cookies, API Keys).
• Architecture Overview
• Agent Usage Guide

---

Run the Server

cd backend python agent_server.py

*Wait until you see: `Agent ready with X tools.`*

### Step 2: Start the Frontend (UI)

Open a **second terminal** and navigate to the frontend folder:

```powershell
cd frontend

# Install dependencies (first time only)
npm install

# Start the Dev Server
npm run dev

Wait until you see: Ready in ... http://localhost:3000

Step 3: Use the System

1. Open http://localhost:3000 in your browser.
2. Chat with AI:
   • Click the "AI Assistant" button (top right).
   • Type: "Search for cat images on Xiaohongshu".
   • The Agent (in Terminal 1) will open Chrome, search, and reply to you in the chat.
3. Publish:
   • Create a design on the canvas.
   • Click Publish -> Xiaohongshu.
   • The Agent will navigate to the publish page and attempt to upload (WIP).

---

🛠️ Development & Debugging

Running the CLI Agent (Auto-Pilot)

If you don't want to use the Web UI, you can run the agent directly in the terminal:

python agent_chrome.py

• Interactive Mode: Type commands manually.
• Auto-Pilot: Type auto to let it run autonomously.
• Pause: Press p to pause execution.

Common Issues

1. "Failed to connect to MCP server"

• Cause: The Node.js bridge path is wrong OR Chrome is not reachable.
• Fix: Check CONFIGURATION_GUIDE.md. Ensure mcp-server-stdio.js path is correct. Try opening Chrome with --remote-debugging-port=9222 manually.

2. "Agent not initialized" (API Error)

• Cause: The Python server failed to connect to the bridge on startup.
• Fix: Check the logs in Terminal 1. Restart agent_server.py.

---

📂 File Structure

xhs-mcp-server/
├── agent_server.py       # 🟢 Backend API (FastAPI)
├── agent_chrome.py       # 🔵 CLI Agent Tool
├── agent_core.py         # 🧠 Agent Logic (LangGraph)
├── session_manager.py    # 🍪 Cookie Injection
├── CONFIGURATION_GUIDE.md # ⚙️ Setup Instructions
├── sns-agent/            # 🎨 Frontend (Next.js)
│   ├── app/              #    React Components
│   └── ...
└── ...