Amish Kushwaha - Apr 17, 2025

Model Context Protocol (MCP) in Action: A Minimal Demo

Before diving into the nitty-gritty of MCP servers and clients, it’s important to witness its power firsthand.

In this post, we’re taking the first tiny yet magical step into the world of MCP — and we’ll do it with a hands-on demo that takes less than 5 minutes.

Ever wondered what it’s like to give your AI access to your actual files — and have it answer smart, contextual questions about them? That’s exactly what we’re about to do.

This is an example showing you how an LLM like Claude can:

Read real files from your computer
Understand logs and CSVs and give insights

All using Model Context Protocol — a way for models to access tools, files, and APIs safely and intelligently.

Requirements:

Node.js installed (comes with npx)
A Claude account + claude desktop app
A folder of files to demo — either your own or from this GitHub repo

Inside, you’ll find :
- server.log — a log file with realistic server errors
- expense.csv — a dummy financial sheet
- claude_config.json — a plug-and-play MCP setup

Step 1: Set Up

Install npx package manager (if you haven’t already)
```
brew install node
npx -v     # verify installation
```
This gives you access to npx, which we’ll use to run the MCP server.

Step 2: Prepare the Demo Files

Download the sample files from the above github repo (or use your files)
Copy the path to this folder

Step 3: Add the Claude Config

Open Claude.ai desktop app
Go to Claude → Settings → Developer → Edit claude_desktop_config.json
Add the provided claude config (replace path to your folder):

{
  &#34;mcpServers&#34;: {
    &#34;filesystem&#34;: {
      &#34;command&#34;: &#34;npx&#34;,
      &#34;args&#34;: [
        &#34;-y&#34;,
        &#34;@modelcontextprotocol/server-filesystem&#34;,
        &#34;/path/to/mcp-demo&#34;
      ]
    }
  }
}

Restart the Claude desktop app

This config tells Claude to:

• Launch a local filesystem-based MCP server

• Allow it to safely access the specified folder

Screenshot Image

Note: Look for this hammer icon to confirm MCP tools are active.

Step 4: Ask some questions

Example:

What errors are there in the log file?
Why did the request to /dashboard fail?
What was the total spend in January?
Which category had the highest expense?

Claude will read the files in real-time and respond with useful, structured answers.

Why This Matters

This demo is intentionally simple. It’s not about building a full-fledged agent — it’s about experiencing how easy and powerful MCP can be.

With just one command and one config file, you’ve given an AI the ability to:

Use tools
Read structured and unstructured files
Reason over context
Provide rich insights

That’s Agentic AI — in its purest form.

Need a walkthrough? Contact your Customer Success Manager or email us at info@bluefunda.com.