RAG MCP Server & Next.js Frontend

A full-stack Retrieval-Augmented Generation (RAG) application built with a Model Context Protocol (MCP) server backend and a modern Next.js frontend. This project enables users to chat with their documents (PDFs) using OpenAI's models and a vector database for context retrieval.

🚀 Features

• MCP Server Backend (Node.js/TypeScript)

  • Implements the Model Context Protocol (MCP).
  • Document Ingestion: Parses PDFs using pypdf (Python bridge) and chunking.
  • Vector Search: utilizing Qdrant for high-performance similarity search.
  • Embeddings: OpenAI text-embedding-3-small.
  • Tools: ingest_document (for admin/setup) and query_knowledge_base (for RAG).

• Next.js Frontend

  • Chat Interface: Clean, responsive UI built with Tailwind CSS and shadcn/ui.
  • Streaming: Real-time token streaming using Vercel AI SDK (v6).
  • Input Management: Robust handling of user inputs and tool invocations.

📂 Project Structure

├── backend/                  # MCP Server Implementation
│   ├── src/
│   │   ├── lib/              # OpenAI & Qdrant clients
│   │   ├── tools/            # MCP Tools (ingest.ts, query.ts)
│   │   ├── mcp-server.ts     # Main server entry point
│   │   └── index.ts          # Server startup
│   ├── scripts/
│   │   ├── manual_ingest.ts  # Script to manually trigger ingestion
│   │   └── parse_doc.py      # Python script for PDF parsing
│   ├── data/                 # Directory for storing raw documents
│   └── package.json
│
├── frontend/                 # Next.js App Router Application
│   ├── src/
│   │   ├── app/
│   │   │   ├── api/chat/     # Route Handler for AI SDK
│   │   │   └── page.tsx      # Main Chat UI Component
│   │   └── components/       # UI Components (shadcn/ui)
│   └── package.json
└── README.md

🛠️ Prerequisites

• Node.js (v18 or higher)
• Python (v3.10+) with pip
• Docker (Desktop or Engine) for running Qdrant
• OpenAI API Key

⚡ Setup Instructions

1. Clone the Repository

git clone <your-repo-url>
cd <repo-name>

2. Backend Setup

1. Navigate to the backend:

   cd backend
   npm install

2. Install Python dependencies (for PDF parsing):

   pip install pypdf
   # If pip fails, try: python -m pip install pypdf

3. Configure Environment Variables: Copy .env.example to .env and fill in your keys.

   OPENAI_API_KEY=sk-...  # Your OpenAI Key
   QDRANT_URL=http://localhost:6333
   PORT=3001
   MODE=sse

3. Frontend Setup

1. Navigate to the frontend:

   cd frontend
   npm install

2. Configure Environment Variables: Create .env.local:

   OPENAI_API_KEY=sk-...  # Required for AI SDK
   MCP_SERVER_URL=http://localhost:3001/sse

4. Start the Vector Database (Qdrant)

You must have Docker running.

docker run -d -p 6333:6333 -v $(pwd)/qdrant_storage:/qdrant/storage:z qdrant/qdrant

Note: If you encounter an error starting Docker, ensure Docker Desktop is running.

5. Ingest Documents

Place your PDF files in backend/src/tools/PDF/ (or update the path). Run the manual ingestion script:

cd backend
npx tsx scripts/manual_ingest.ts

6. Run the Application

You need to run both servers (use two terminals).

Terminal 1 (Backend):

cd backend
npm run dev

Terminal 2 (Frontend):

cd frontend
npm run dev

Access the chat interface at http://localhost:3000.

CURRENT ISSUES: 1)The chatbot front interface is working but not being able to retrieve data and access the embeddings. Currently trying to fix this issue. 2)- RAG Connection: Accessing the query_knowledge_base tool depends on a local Qdrant instance. If Qdrant is not running (e.g., Docker issues), the chat will fall back to general knowledge.