FAQ‑Bot – ChromaDB + MCP Tool

Overview

This project implements an FAQ‑bot that answers questions about local course notes stored in Markdown files using ChromaDB for retrieval and a single MCP‑style tool for metadata queries.

• Local knowledge – 2–3 .md files in data/.
• Vector store – ChromaDB with Ollama embeddings.
• Tool – fetch_course_meta that performs a GET request to a local JSON file (mock MCP server).
• Agent – LangGraph agent that routes the query to the appropriate source and labels the answer with source: chroma | mcp_meta.
• CLI – 3 example questions (2 for Chroma, 1 for meta) and an interactive mode.

Setup

# Pull Ollama model
ollama pull nomic-embed-text

# Install dependencies
pip install -r requirements.txt

# Start local mock MCP server (JSON file)
python -m http.server 8000 --directory mock

Running the Bot

python main.py --question "What is the course schedule?"

Example Questions

1. Chroma – "What topics are covered in week 3?"
2. Chroma – "How many hours per week are required?"
3. Meta – "What is the course schedule?"

Project Structure

├── data/                 # Markdown FAQ files
├── mock/                 # JSON mock MCP server
├── src/
│   ├── agent.py          # LangGraph agent definition
│   ├── chunker.py        # Chroma loader
│   ├── cli.py            # CLI interface
│   ├── config.py         # Config & constants
│   ├── init_loader.py    # Bulk loader
│   ├── qdrant_store.py   # Vector store helper
│   ├── rag_tools.py      # Search & meta tools
│   ├── tools.py          # MCP wrapper
│   └── vector_store.py   # Chroma store
└── requirements.txt

License

MIT