database_mcp_agent

Database Assistant powered by Agno and GibsonAI MCP server

This project implements a conversational AI agent that acts as a GibsonAI database assistant. It can help you manage your database projects and schemas using natural language.

Features

• Create GibsonAI Projects: Start new database projects from scratch.
• Schema Management: Define and modify tables, columns, and relationships.
• Deploy Changes: Apply schema changes to your hosted databases.
• Querying: Interact with your database schema and data.
• Best Practices: Get insights and recommendations on database structure.

How It Works

The agent is built using the agno framework. It uses the meta-llama/Meta-Llama-3.1-70B-Instruct model hosted on Nebius for language understanding and generation. The MCPTools from the agno library provide a bridge to the gibson-cli, allowing the agent to execute commands and interact with GibsonAI.

Getting Started

Prerequisites

• Python 3.8+
• An account with GibsonAI
• A Nebius API key

Installation

1. Clone the repository:

   git clone https://github.com/your-username/awesome-llm-apps.git
   cd awesome-llm-apps/mcp_ai_agents/database_mcp_agent

2. Set up a virtual environment and install dependencies: This project uses uv for package management.

   uv venv
   source .venv/bin/activate
   uv pip install -r requirements.txt 

   (Note: Assuming a requirements.txt will be created. The pyproject.toml is present, so uv pip install . might be more appropriate if the project is packaged correctly. For now, I'''ll assume a simple setup)

3. Configure your environment variables: Create a .env file in the database_mcp_agent directory and add your Nebius API key:

   NEBIUS_API_KEY="your_nebius_api_key"
   

Usage

To run the agent, execute the main.py script:

python main.py

You can modify the message in the if __name__ == "__main__": block in main.py to change the initial prompt for the agent. For example:

if __name__ == "__main__":
    asyncio.run(
        run_gibsonai_agent(
            '''
            Create a new GibsonAI project for my e-commerce store.
            It should have tables for products, customers, and orders.
            '''
        )
    )

How to get gibson-cli?

To install the gibson-cli, you can use uvx or pipx:

uvx --from gibson-cli@latest gibson --version

This will download and run the Gibson CLI without installing it globally. The agent uses a similar command to interact with the MCP server.