MCP-Based RAG System for Semiconductor Component Search
This project demonstrates MCP (Model Context Protocol) integration with ChromaDB and HuggingFace models for Retrieval-Augmented Generation (RAG).
Project Overview
This system shows how MCP works and its purpose:
- MCP provides a standardized protocol for context retrieval
- ChromaDB stores and retrieves semantic embeddings
- RAG Pipeline combines retrieval with LLM generation
- Backend API allows document upload and question-answering
Architecture
User Question → API Endpoint → RAG Pipeline
↓
Retrieval from ChromaDB (via embeddings)
↓
LLM generates answer with context
↓
Response to user
Key Components
- MCP Server (
mcp_server.py): Demonstrates MCP protocol for structured context retrieval - RAG Pipeline (
rag_pipeline.py): Handles embeddings (encoding) and LLM (decoding) - FastAPI Backend (
main.py): REST API for document upload and Q&A - ChromaDB: Vector database for semantic search
Installation
- Install dependencies:
pip install -r requirements.txt
-
Set environment variables: Create a
.envfile or use the provided HF API key inconfig.py -
Create example Excel file:
python create_example_excel.py
Usage
1. Start the API Server
python main.py
The API will be available at http://localhost:8000
2. Upload Excel Document
curl -X POST "http://localhost:8000/upload" \
-H "accept: application/json" \
-F "file=@examples/semiconductor_components.xlsx"
Or use the FastAPI docs at http://localhost:8000/docs
3. Ask Questions
curl -X POST "http://localhost:8000/ask" \
-H "Content-Type: application/json" \
-d '{"question": "What MOSFET components are available?", "n_results": 3}'
API Endpoints
GET /- API informationGET /health- Health checkPOST /upload- Upload Excel documentPOST /ask- Ask a questionGET /info- Get collection information
How MCP Works
MCP (Model Context Protocol) serves as a standardized interface for:
- Context Retrieval: Structured way to query and retrieve relevant information
- Tool Definition: Clear specification of available operations
- Protocol Communication: Standardized communication between components
In this project:
- MCP server defines tools for querying ChromaDB
- RAG pipeline uses MCP principles for context retrieval
- Backend integrates MCP concepts for document processing
Models Used
- Encoding (Embeddings):
sentence-transformers/all-MiniLM-L6-v2 - Decoding (LLM): Llama model from HuggingFace (or fallback to GPT-2)
Example Questions
- "What MOSFET components are available?"
- "Show me voltage regulators from Texas Instruments"
- "What components work with 5V?"
- "List all temperature sensors"
Project Structure
MCP2/
├── main.py # FastAPI backend
├── rag_pipeline.py # RAG pipeline with embeddings & LLM
├── mcp_server.py # MCP server for ChromaDB
├── config.py # Configuration
├── create_example_excel.py # Generate example data
├── requirements.txt # Dependencies
├── examples/ # Example Excel files
└── chroma_db/ # ChromaDB storage (created automatically)
Notes
- First run will download models from HuggingFace (requires API key)
- ChromaDB data persists in
./chroma_db/directory - Uploaded files are stored in
./uploads/directory