🤖 Modular FAQ Agent with MCP & Memory

An agentic FAQ system using LangGraph, Qwen 3, and Qdrant. Implements Model Context Protocol (MCP) for tool use and stateful memory. Modular, containerized, and ready for deployment.

🚀 Features

• Semantic Search: Uses nomic-embed-text embeddings with Qdrant for accurate FAQ retrieval.
• Conversational Memory: Built with LangGraph's MemorySaver to track session-based conversation context.
• MCP Integration: Tools are decoupled from the agent through a standard Model Context Protocol server.
• Streaming Response: Optional SSE support for real-time interaction.
• Containerized Architecture: Full stack deployment using Docker Compose.

📁 Project Structure

mcp_faq_agent/
├── API/                 # FastAPI routes and Pydantic models
│   ├── core/            # Business logic (get_content filtering)
│   └── endpoints.py     # Main API endpoints
├── agent/               # LangGraph agent orchestration
├── mcp_custom/          # MCP Server hosting search tools
├── qdrant_ingestion/    # Scripts to populate the vector store
├── configurations/      # Settings, environment loading, and YAML prompts
└── utils/               # Shared helper functions

🛠️ Getting Started

1. Prerequisites

• Docker & Docker Compose
• (Optional) uv for local development

2. Deployment

Simply build and start the entire stack:

sudo docker compose up -d --build

This will spin up:

• ollama-qwen-cpu: The local LLM engine.
• qdrant: The vector database.
• faq-api: The FastAPI agent entry point.

3. Ingesting Data (Initial Setup)

To populate the database with your specific FAQs:

# Run the ingestion script (local context)
uv run python qdrant_ingestion/ingest_faqs.py

📡 API Usage

POST /chat

Body:

{
  "question": "What is the company remote work policy?",
  "session_id": "user_123"
}

Example Curl:

curl -X POST http://localhost:8000/chat \
     -H "Content-Type: application/json" \
     -d '{"question": "What are the common benefits?", "session_id": "demo_1"}'

⚙️ Configuration

The system uses environment variables managed in .env (passed via Docker Compose):

• LLM_MODEL: The Ollama model to use.
• OLLAMA_URL: Connection URL for Ollama.
• QDRANT_URL: Connection URL for Qdrant.
• PROMPT_PATH: Location of the YAML prompt versioning file.

📄 License

MIT