MCP From Zero: Quick Data

Purpose: Learn to build Powerful Model Context Protocol (MCP) servers by scaling tools into reusable agentic workflows (ADWs aka Prompts w/tools).

Quick-Data

Quick-Data is a MCP server that gives your agent arbitrary data analysis on .json and .csv files.

We use quick-data as a concrete use case to experiment with the MCP Server elements specifically: Prompts > Tools > Resources.

See quick-data-mcp for details on the MCP server

Leading Questions

We experiment with three leading questions:

1. How can we MAXIMIZE the value of custom built MCP servers by using tools, resources, and prompts TOGETHER?
2. What's the BEST codebase architecture for building MCP servers?
3. Can we build an agentic workflow (prompt w/tools) that can be used to rapidly build MCP servers?

Understanding MCP Components

MCP servers have three main building blocks that extend what AI models can do:

Tools

What: Functions that AI models can call to perform actions.

When to use: When you want the AI to DO something at a low to mid atomic level based on your domain specific use cases.

Example:

@mcp.tool()
async def create_task(title: str, description: str) -> dict:
    """Create a new task."""
    # AI can call this to actually create tasks
    return {"id": "123", "title": title, "status": "created"}

Resources

What: Data that AI models can read and access.

When to use: When you want the AI to READ information - user profiles, configuration, status, or any data source.

Example:

@mcp.resource("users://{user_id}/profile")
async def get_user_profile(user_id: str) -> dict:
    """Get user profile by ID."""
    # AI can read this data to understand users
    return {"id": user_id, "name": "John", "role": "developer"}

Prompts

What: Pre-built conversation templates that start specific types of discussions.

When to use: When you want to give the AI structured starting points for common, repeatable workflows for your domain specific use cases.

Example:

@mcp.prompt()
async def code_review(code: str) -> str:
    """Start a code review conversation."""
    # AI gets a structured template for code reviews
    return f"Review this code for security and performance:\n{code}"

Quick Decision Guide

• Need AI to take action? → Use Tools
• Need AI to read data? → Use Resources
• Need Reusable Agentic Workflows (ADWs)? → Use Prompts

Quick Setup

To use the Quick Data MCP server:

1. Navigate to the MCP server directory:

   cd quick-data-mcp/

2. Configure for your MCP client:

   # Copy the sample configuration
   cp .mcp.json.sample .mcp.json
   
   # Edit .mcp.json and update the --directory path to your absolute path
   # Example: "/Users/yourusername/path/to/quick-data-mcp"

3. Test the server:

   uv run python main.py

See quick-data-mcp/README.md for complete setup and usage documentation.

Resources

• MCP Clients: https://modelcontextprotocol.io/clients
• Claude Code Resource Support Github Issue: https://github.com/anthropics/claude-code/issuesç/545

Master AI Coding

Learn to code with AI with foundational Principles of AI Coding

Follow the IndyDevDan youtube channel for more AI coding tips and tricks.

Use the best Agentic Coding Tool: Claude Code