Content
# Tool List
> **SHEIN Fashion PLM + Multi-Agent + MCP Protocol Complete Technical Solution and Open Source Demo**
[](LICENSE)
[](https://python.org)
[](https://go.dev)
[](https://langchain-ai.github.io/langgraph/)
[](https://modelcontextprotocol.io)
## Project Overview
Using **Multi-Agent** to streamline the entire fashion PLM process, completing the automation of "trend collection → product selection → style generation → cost verification → hot sales review → data feedback iteration model".
**Core capabilities: replacing 70% of manual repetitive work, increasing new product hit rate by 30%+.**
## Technical Architecture
```
┌─────────────────────────────────────────────┐
│ Front-end Business Application Layer │
│ Go Dashboard · React AI Panel · Celery Queue │
├─────────────────────────────────────────────┤
│ Multi-Agent Orchestration Layer (Core) │
│ LangGraph · 5 types of Agents · Three-layer Memory · AutoGen │
├─────────────────────────────────────────────┤
│ MCP Standardized Tool Layer │
│ Database MCP · Multimodal MCP · Crawler MCP · PLM Sync MCP │
├─────────────────────────────────────────────┤
│ Underlying Data & Model Computing Layer │
│ PostgreSQL · Milvus · DeepSeek · Data Governance │
└─────────────────────────────────────────────┘
```
## Quick Start
```bash
# 1. Clone the project
git clone https://github.com/YOUR_USER/plm-ai-agent.git
cd plm-ai-agent
# 2. Install dependencies
pip install -r requirements.txt
# 3. Start GUI (no API Key required to run Demo)
python server.py
# 4. Open browser
open http://localhost:8765
```
## Web GUI Demo
| Function | Description |
|------|------|
| Left Panel | Input product selection requirements + cost cap |
| Real-time Steps | WebSocket pushes each Agent execution status |
| Result Display | Style solution card + cost verification + hit rate prediction |
| Dark Theme | Consistent with PPT scheme document style |
> **No DEEPSEEK_API_KEY required**: Demo uses simulated data to complete the entire process. Set `DEEPSEEK_API_KEY` to enable real LLM inference.
## Technical Architecture
```
┌─────────────────────────────────────────────┐
│ Front-end GUI (HTML/CSS/JS) │
│ WebSocket real-time push Agent execution steps │
├─────────────────────────────────────────────┤
│ FastAPI Backend (server.py) │
│ REST API + WebSocket + Static File Service │
├─────────────────────────────────────────────┤
│ Multi-Agent Orchestration Layer │
│ Supervisor → Trend/PLM Data (parallel) → Design → Supply Chain │
├─────────────────────────────────────────────┤
│ Memory + MCP Tool Layer │
│ Three-layer Memory + 4 types of MCP Server framework │
└─────────────────────────────────────────────┘
```
## CLI Demo
```bash
# Command-line version (terminal interaction)
python demo.py
```
## Project Structure
```
plm-ai-agent/
├── agent/ # Multi-Agent orchestration engine (LangGraph)
│ ├── planner.py # Supervisor planning Agent
│ ├── trend.py # Trend retrieval Agent
│ ├── plm_data.py # PLM data Agent
│ ├── design.py # Multimodal generation Agent
│ └── supply_chain.py # Supply chain verification Agent
├── mcp-servers/ # MCP standardized tool services
│ ├── db_server.py # Database MCP Server
│ ├── multimodal_server.py # Multimodal MCP Server
│ ├── crawler_server.py # Crawler MCP Server
│ └── plm_sync_server.py # PLM sync MCP Server
├── memory/ # Three-layer Memory system
│ ├── working.py # Working memory
│ ├── session.py # Session memory (Redis)
│ └── long_term.py # Long-term memory (Mem0 + Milvus)
├── frontend/ # Front-end Demo
│ ├── dashboard.go # Go monitoring dashboard
│ └── ai-panel/ # React AI product selection panel
├── data/ # Sample data & training set
├── docs/ # Complete technical solution PPT
├── demo.py # Demo entry
└── docker-compose.yml
```
## License
MIT © 2026 <br>
