Enterprise Data Agent 🗄️

A natural-language interface to an enterprise employee database, powered by Groq LLM tool-calling and a local SQLite database, with a Streamlit frontend and a Model Context Protocol (MCP) server backend.

Architecture

flowchart TD
    %% Styling
    classDef client fill:#2d3748,stroke:#94a3b8,stroke-width:1px,color:#f8fafc,rx:10px
    classDef core fill:#0f172a,stroke:#3b82f6,stroke-width:2px,color:#eff6ff,rx:6px
    classDef db fill:#064e3b,stroke:#10b981,stroke-width:2px,color:#ecfdf5,rx:6px
    classDef ai fill:#312e81,stroke:#8b5cf6,stroke-width:2px,color:#eef2ff,rx:6px

    %% Nodes
    User(["👤 User Browser"]):::client
    ExtAI(["🤖 MCP Client (Claude, Cursor)"]):::client

    App["⚡ Streamlit App"]:::core
    Server["🔌 MCP Server"]:::core
    DBLayer["⚙️ Shared DB Layer"]:::core

    Groq{"🧠 Groq LLM"}:::ai
    SQLite[("💾 SQLite Database")]:::db

    %% Connections
    User <-->|"Chat Interface"| App
    App <-->|"Prompts & Tools"| Groq
    App -->|"Execute Tools"| DBLayer

    ExtAI <-->|"MCP / stdio"| Server
    Server -->|"Execute Tools"| DBLayer

    DBLayer <-->|"SQL Operations"| SQLite

Quick Start

1. Prerequisites

Python 3.11+
A Groq API key

2. Clone & Set Up

git clone <your-repo-url>
cd mcp_project

# Create and activate a virtual environment
python -m venv venv
venv\Scripts\activate        # Windows
# source venv/bin/activate   # macOS/Linux

# Install dependencies
pip install -r requirements.txt

3. Configure Environment

Create a .env file in the project root (F:\PROJECTS\MCP\.env):

GROQ_API_KEY=gsk_your_key_here
GROQ_MODEL=llama-3.3-70b-versatile   # optional, this is the default

4. Initialize the Database

# From the project root (F:\PROJECTS\MCP\)
sqlite3 enterprise_data.db < schema.sql

5. Run the App

# From F:\PROJECTS\MCP\
streamlit run mcp_project/app.py

Open http://localhost:8501 in your browser.

Available Tools

The LLM can call these tools automatically based on your query:

Tool	Description
`get_active_users`	List all active users and their roles
`add_user`	Add a new user (name + role)
`deactivate_user`	Deactivate an active user by name
`get_user_stats`	Get total, active, and inactive counts
`search_users`	Search users by name fragment

Example Queries

"Who are our active users?"
"Add Jane Doe as a Manager"
"How many users do we have in total?"
"Search for users named Alice"
"Deactivate Bob Jones"
"Show me user statistics"

Running the MCP Server (Standalone)

server.py exposes the same tools via the Model Context Protocol. It supports both local stdio connections and an HTTP-based sse Transport!

It now exposes:

Tools: DB query and update operations
Resources: Exposes schema.sql over MCP
Prompts: Provides pre-configured prompts like hr-assistant

To run locally for AI desktop apps (like Claude Desktop / Cursor):

python mcp_project/server.py --transport stdio

To run as an HTTP SSE server on port 8000:

python mcp_project/server.py --transport sse --port 8000

Project Structure

F:\PROJECTS\MCP\
├── .env                    # ← API key (never commit this!)
├── .gitignore
├── requirements.txt
├── schema.sql              # DB schema — use to recreate the database
├── enterprise_data.db      # SQLite database (gitignored)
└── mcp_project\
    ├── app.py              # Streamlit frontend
    ├── server.py           # MCP stdio server
    └── db.py               # Shared database access layer

Security Notes

Never commit your .env file. The .gitignore excludes it.
Never commit enterprise_data.db. Use schema.sql to recreate it.
Rotate your Groq API key at console.groq.com if it has ever been exposed.

Enterprise Data MCP Server