Enterprise Data Agent

An enterprise-grade, secure, and privacy-first Autonomous Data Agent built using the Model Context Protocol (MCP), FastMCP, LlamaIndex Workflows, and a localized LLaMA 3.2 model via Ollama. This agent acts as an automated SQL assistant that interacts securely with an on-premise PostgreSQL database using dynamic tool-calling capability, ensuring no database schemas are exposed raw to the external world.

🚀 Key Features

• Model Context Protocol (MCP): Implements modern 2026 standardized client-server architecture (FastMCP) over Server-Sent Events (SSE).
• Privacy-First Architecture: Utilizes local llama3.2:1b for zero-data leak enterprise compliances.
• Dynamic Tool Calling: Built-in SQL execution layer protecting database context via strict system prompting (list_tables, read_data, add_data).
• Streamlit User Interface: A production-style, dynamic chat interface supporting streaming statuses of agent thoughts and backend tool invocations.

🛠️ Tech Stack

• AI Framework: LlamaIndex (FunctionAgent Workflows)
• MCP Server Framework: FastMCP (Python)
• Database Driver: Psycopg3 (Modern PostgreSQL)
• Local LLM Engine: Ollama (LLaMA 3.2 1B)
• Frontend UI: Streamlit

📁 Project Structure

• server.py - The standalone FastMCP server exposing database query and schema capabilities securely.
• agent_notebook.ipynb - Core testing and modular workflow pipeline using LlamaIndex client specs.
• app.py - Production-ready UI frontend wrapping the async agent loop.

🏃 How to Run

Step 1: Start the FastMCP Server

Ensure your local PostgreSQL database is up and matching the config, then run:

python server.py --server_type sse --port 8000