MCP Talk Demo Files

This folder contains practical examples for the "Beyond the Basics: Unlocking the Power of MCP" presentation at the 10x Developers Meetup.

🚀 Quick Setup (Recommended)

Option 1: Automated Setup (Recommended)

macOS/Linux:

cd /Users/davidhague/source/mcp-talk
chmod +x setup.sh
./setup.sh

Windows:

cd \Users\davidhague\source\mcp-talk
setup.bat

Option 2: Manual Setup

cd /Users/davidhague/source/mcp-talk

# Create and activate virtual environment
python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install --upgrade pip
pip install -r requirements.txt

# Test setup
python test_setup.py

📁 Demo Files

1. `01_python_client_demo.py`

Simple MCP Python Client

• Demonstrates how to connect to any MCP server from Python
• Shows tool discovery, resource access, and general MCP interactions
• Perfect for showing the developer experience

To run:

source venv/bin/activate  # Activate venv first!

# Terminal 1 - Start MCP server
npx -y @modelcontextprotocol/server-filesystem /tmp

# Terminal 2 - Run client 
python 01_python_client_demo.py

2. `02_system_monitor_server.py`

Complete MCP Server from Scratch

• System monitoring server with all three MCP capabilities
• Tools: CPU usage, memory info, disk usage, processes, network
• Resources: System overview, platform information
• Shows how to build a full MCP server

To run:

source venv/bin/activate
python 02_system_monitor_server.py

To connect with Claude Desktop:

Add to your Claude Desktop config (`~/Library/Application Support/Claude/claude_desktop_config.json`):

{
  "mcpServers": {
    "system-monitor": {
      "command": "python",
      "args": ["/Users/davidhague/source/mcp-talk/02_system_monitor_server.py"],
      "env": {
        "PATH": "/Users/davidhague/source/mcp-talk/venv/bin:/usr/bin:/bin"
      }
    }
  }
}

3. `03_sse_server_complete.py`

MCP with Server-Sent Events (SSE) - Pseudocode Reference

• LEGACY TRANSPORT - Educational purposes only
• Shows SSE endpoint structure (GET /sse + POST /messages)
• Demonstrates why Streamable HTTP is preferred

Note: This is pseudocode for understanding SSE concepts.

For working implementations, use `04_https_streamable.py`

4. `04_https_streamable.py`

Next-Generation Streamable HTTP MCP

• Demonstrates the latest MCP transport protocol
• Unified endpoint architecture
• Production-ready streaming implementation

To run:

source venv/bin/activate
python 04_https_streamable.py

5. `05_http_client_demo.py`

HTTP MCP Client Demo

• Demonstrates connecting to HTTP MCP servers instead of stdio
• Shows the same LLM integration but over HTTP transport
• Perfect companion to show stdio vs HTTP client differences

To run:

source venv/bin/activate

# Terminal 1 - Start HTTP MCP server
python 04_https_streamable.py

# Terminal 2 - Run HTTP client
python 05_http_client_demo.py

🎯 Presentation Flow

Demo 1: Python Client (stdio) (5 minutes)

• "Here's how easy it is to connect to any MCP server from Python"
• Run the stdio client against filesystem server
• Show tool discovery and execution with LLM integration

Demo 2: Build Custom Server (10 minutes)

• "Let's build an MCP server from scratch"
• Walk through the system monitor server code
• Show tools and resources capabilities
• Connect it to Claude Desktop

Demo 3: HTTP Transport (8 minutes)

• "Now let's make servers truly standalone with HTTP"
• Run the HTTP streamable server
• Show the unified endpoint architecture
• Demonstrate HTTP client connecting to it

Demo 4: SSE Concepts (2 minutes)

• "For context, here's the legacy SSE approach"
• Show the pseudocode structure
• Explain why Streamable HTTP is preferred

🔧 Key Technical Points

MCP Architecture

┌─────────────┐    MCP Protocol    ┌─────────────┐
│   Client    │ ◄─────────────────► │   Server    │
│ (Claude,    │                     │ (Your App)  │
│  Cursor,    │   • Tools           │             │
│  Custom)    │   • Resources       │             │
└─────────────┘   • Prompts         └─────────────┘

Transport Evolution

1. STDIO → Local, simple, process-based

2. SSE → Remote, dual-endpoint (GET /sse + POST /messages)

3. Streamable HTTP → New unified, single endpoint

Three Capabilities

• Tools (Model-controlled): Functions the AI can call
• Resources (App-controlled): Data the AI can read
• Prompts (User-controlled): Templates users select

🐍 Virtual Environment Notes

Why use a virtual environment?

• Isolated Python dependencies
• Reproducible demo environment
• No conflicts with system packages
• Professional development practice

Managing the venv:

# Activate (do this before running demos)
source venv/bin/activate

# Check it's working
which python  # Should show venv/bin/python

# Install additional packages if needed
pip install some-package

# Deactivate when done
deactivate

If you see import errors:

• Make sure venv is activated (`source venv/bin/activate`)
• Run `python test_setup.py` to verify setup
• Reinstall with `pip install -r requirements.txt`

📚 Additional Resources

• MCP Documentation
• MCP Python SDK
• MCP Servers Collection
• MCP Inspector - Great for testing

🚀 Next Steps for Audience

1. Start Simple: Use existing MCP servers with Claude Desktop

2. Build Custom: Create MCP servers for your specific needs

3. Go Remote: Deploy SSE servers for team/production use

4. Explore Transports: Try different transport protocols for your use case

💡 Talk Takeaways

• MCP is the "USB-C for AI" - universal protocol
• Three capabilities: tools, resources, and prompts unlock real workflows
• Multiple transport options for different deployment scenarios
• Production-ready with proper session management
• Open standard with growing ecosystem

🔧 Troubleshooting

Common Issues:

1. ModuleNotFoundError: Activate venv first (`source venv/bin/activate`)

2. Permission denied on setup.sh: Run `chmod +x setup.sh`

3. Node.js not found: Install from nodejs.org for filesystem server demo

4. Port 8000 in use: Change port in SSE server or kill existing process

5. Claude Desktop not connecting: Check config file path and restart Claude Desktop

*These demos show practical, working examples of MCP's power beyond basic tutorials.*