Agentic RAG Chatbot with MCP

An advanced Retrieval-Augmented Generation (RAG) chatbot featuring an agent-based architecture using Model Context Protocol (MCP) for seamless communication between components. This implementation uses FastAPI for the backend and includes a modular agent system for processing and responding to user queries.

🚀 Features

• Multi-format Document Support: Process PDF, DOCX, PPTX, CSV, and TXT files
• Agent-Based Architecture: Modular design with specialized agents for ingestion, retrieval, and response generation
• Model Context Protocol (MCP): Standardized communication protocol between agents
• Vector Search: Semantic search capabilities using FAISS and sentence-transformers
• Web Interface: Interactive UI for document uploads and chat
• Local-First: Runs entirely on your machine with local models by default
• Asynchronous Processing: Non-blocking operations for better performance

🏗️ Architecture

Core Components

Agents

• IngestionAgent: Processes and chunks uploaded documents, handles text extraction and splitting
• RetrievalAgent: Performs semantic search using sentence-transformers and FAISS
• ResponseAgent: Generates contextual responses using language models
• CoordinatorAgent: Manages workflow and message routing between agents using MCP

Storage

• FAISS Vector Store: Efficient similarity search with all-MiniLM-L6-v2 embeddings (384 dimensions)
• Local File System: Stores uploaded documents in the uploads/ directory

API Endpoints

• POST /api/upload: Upload and process documents
• POST /api/chat: Send chat messages and get responses
• GET /health: Check service status
• POST /api/clear_kb: Clear the knowledge base

🛠️ Installation

Prerequisites

• Python 3.10+
• pip (Python package manager)
• Git

Setup

1. Clone the repository:

   git clone https://github.com/yourusername/agentic-chatbot.git
   cd agentic-chatbot

2. Create and activate a virtual environment:

   python -m venv venv
   # Windows:
   venv\Scripts\activate
   # Unix/MacOS:
   source venv/bin/activate

3. Install dependencies:

   pip install -r requirements.txt

   Note: The first run will download the all-MiniLM-L6-v2 model (~80MB)

4. Set up environment variables:

   cp .env.example .env

   The default configuration in .env is pre-configured for local development:

   # Server
   HOST=0.0.0.0
   PORT=8000
   DEBUG=true
   
   # Models
   EMBEDDING_MODEL=all-MiniLM-L6-v2
   LLM_MODEL=distilgpt2
   
   # Storage
   UPLOAD_FOLDER=uploads
   
   # Optional: Set to "cuda" if you have a CUDA-compatible GPU
   # DEVICE=cuda

🚀 Running the Application

1. Start the FastAPI server with auto-reload for development:

   uvicorn main:app --reload

2. Access the application:

   • Web Interface: http://localhost:8000
   • API Documentation: http://localhost:8000/docs
   • Health Check: http://localhost:8000/health

🖥️ Usage

Uploading Documents

1. Using the Web Interface:

   • Navigate to http://localhost:8000
   • Click "Upload Document"
   • Select a file (PDF, DOCX, PPTX, CSV, or TXT)
   • Wait for processing to complete

2. Using the API:

   curl -X 'POST' \
     'http://localhost:8000/api/upload' \
     -H 'accept: application/json' \
     -H 'Content-Type: multipart/form-data' \
     -F 'file=@your_document.pdf;type=application/pdf'

Chat Interface

1. Web Interface:

   • Type your question in the input box
   • Press Enter or click "Send"
   • View the response with relevant document sources

2. API Usage:

   curl -X 'POST' \
     'http://localhost:8000/api/chat' \
     -H 'accept: application/json' \
     -H 'Content-Type: application/json' \
     -d '{
       "message": "What is this document about?",
       "conversation_id": "optional_conversation_id"
     }'

⚙️ Configuration

The application can be configured using environment variables in the .env file:

# Server Configuration
HOST=0.0.0.0  # Bind address
PORT=8000      # Port to run the server on
DEBUG=true     # Enable debug mode (not recommended for production)

# Model Configuration
EMBEDDING_MODEL=all-MiniLM-L6-v2  # Sentence transformer model for embeddings
LLM_MODEL=distilgpt2              # Language model for response generation

# Storage Configuration
UPLOAD_FOLDER=uploads  # Directory to store uploaded files

# Hardware Configuration
# DEVICE=cpu  # Set to "cuda" if you have a CUDA-compatible GPU

# Logging
LOG_LEVEL=INFO  # DEBUG, INFO, WARNING, ERROR, CRITICAL
VECTOR_STORE_PATH=vector_store.faiss

📂 Project Structure

agentic-chatbot/
├── main.py               # Main application and web server
├── document_processor.py # Document parsing and text extraction
├── embedding_service.py  # Text embeddings and similarity search
├── agents.py            # Agent system with MCP implementation
├── minimal_requirements.txt  # Python dependencies
├── uploads/             # Directory for uploaded files
└── README.md

Troubleshooting

Common Issues

1. Model Loading Issues:

   • Ensure you have enough disk space for the models
   • Check your internet connection if downloading models for the first time
   • Verify the model names in your .env file are correct

2. Document Processing Failures:

   • Make sure the uploaded files are not corrupted
   • Check that the file formats are supported
   • Verify file permissions in the uploads directory

3. Performance Issues:

   • The application runs locally and may be slow on less powerful machines
   • Consider using a smaller model if performance is a concern

License

MIT License - see the LICENSE file for details.

Acknowledgements

• FastAPI - Modern Python web framework
• Hugging Face - Transformers and models
• sentence-transformers - For text embeddings