sqlite-mcp-agent

Local MCP Client Demo

This project demonstrates a fully local Model Context Protocol (MCP) client using:

• SQLite via Python sqlite3
• FastMCP (mcp-server) for tool management
• DeepSeek-R1 LLM hosted locally via Ollama
• LlamaIndex for building an MCP-powered agent

📦 Requirements

Ensure you have the following installed:

• Python 3.8+

• Ollama CLI (for DeepSeek-R1)

• System packages:

  pip install -r requirements.txt
  

🛠️ Files Overview

• server.py — MCP server exposing add_data & read_data tools backed by SQLite
• ollama_client.py — Async agent setup using Ollama + LlamaIndex
• chat_interface.py — CLI wrapper to chat with the agent
• requirements.txt — Python dependencies

🚀 Setup & Running

1. Start the MCP Server

   # Initialize database and run in SSE mode
   python server.py --server_type sse
   

2. Launch Ollama & DeepSeek-R1

   ollama pull deepseek-r1
   ollama run deepseek-r1
   

3. Install Python Dependencies

   pip install -r requirements.txt
   

4. Run the Chat Client

   python chat_interface.py
   

   Type your queries (e.g., add_data(...) or natural questions) and see the agent invoke tools and respond.

🎯 Best Practices

• Tool Design: Keep tools focused (single responsibility) and clearly document inputs/outputs.
• Error Handling: Catch and log exceptions in tool implementations to avoid silent failures.
• Security: Avoid executing raw SQL when possible; use parameterized queries for production.
• Extensibility: Add new MCP tools by decorating with @mcp.tool() and exposing via list_tools.

📝 Next Steps

• Integrate additional data sources (e.g., external APIs).
• Deploy on Lightning AI for scalable hosting.
• Enhance agent prompts and caching strategy for performance.

Demo built with ❤️ for fully local, context-aware agents!