🚨 MCP Disaster Alert Center

An intelligent AI-powered Model Context Protocol (MCP) server and client demonstrating real-time disaster alert management with LLM-based tool orchestration.

🎯 Overview

This project showcases a disaster alert command center built with:

• MCP Server (.NET 10) - Provides disaster alert tools via JSON-RPC 2.0 over WebSocket
• AI Agent Client (.NET 10) - Natural language interface with streaming responses
• Google Gemma 4 LLM - Intelligent tool selection and reasoning
• Real-time Alerts - Simulated disaster events streaming to subscribed clients

✨ Features

🤖 AI Agent Architecture

• Natural Language Understanding - Accept free-form user input
• Agent Reasoning - LLM explains its understanding in human terms
• Intelligent Routing - Selects appropriate tool based on intent
• Fallback Matching - Keyword matching when LLM unavailable
• Streaming Output - ChatGPT/Claude-style character-by-character streaming

🛠️ Available Tools

1. subscribe_disaster - Subscribe to alert types (earthquake, flood, storm)
2. get_alerts - Retrieve all active disaster incidents
3. incident_summary - AI-powered analysis of current incidents

🚀 Quick Start

Prerequisites

• .NET 10 SDK
• Google Gemma 4 API key (get at Google AI Studio)

Installation

# Clone the repository
git clone https://github.com/ndoteddy/mcp-disaster-center.git
cd mcp-disaster-center

# Set up API key (PowerShell)
$env:GEMMA_API_KEY = "your-api-key-here"

Running the Demo

Terminal 1 - Start the MCP Server:

cd server
dotnet run -c Debug

Terminal 2 - Start the AI Agent Client:

cd client
dotnet run -c Debug

Example Usage

🧠 You: show me active incidents
🤖 Agent: Thinking...
💭 Agent Reasoning: The user wants to see a list of current disaster 
    incidents to stay informed about ongoing emergencies.
🔧 Selected tool: get_alerts
✔️ [10/10] FLOOD in Hilltown @ 08:08:41
   [8/10] EARTHQUAKE in Coastville @ 08:08:43
   [6/10] STORM in Hilltown @ 08:07:27

Commands

/help              - Show all commands
/tools             - List available tools
/subscribe <type>  - Subscribe to alerts (earthquake, flood, storm)
/get-alerts        - Get current active alerts
/incident-summary  - AI analysis of active incidents
/quit              - Exit

🏗️ Architecture

MCP Protocol

• JSON-RPC 2.0 compliant messaging
• WebSocket transport for real-time communication
• Tool discovery via tools/list method
• Tool invocation via tools/call method
• Real-time notifications for subscribed clients

AI Agent Flow

1. User Input - Natural language command
2. Agent Reasoning - LLM explains understanding
3. Tool Selection - LLM chooses appropriate tool
4. Execution - Invoke tool via MCP protocol
5. Results - Stream response character-by-character

Components

Server (MCP Protocol Provider)

• WebSocket listener on localhost:5000
• Tool definitions for disaster management
• Real-time alert generation (2-5 second intervals)
• LLM integration for incident analysis
• Session/subscription management

Client (AI Agent)

• Tool discovery on startup
• Dual-stage LLM reasoning
• Graceful fallback to keyword matching
• Streaming character-by-character output
• Error resilience and recovery

🔌 API Integration

Google Gemma 4

• Endpoint: https://generativelanguage.googleapis.com/v1beta/models/gemma-4-31b-it:generateContent
• Auth: Query parameter key={GEMMA_API_KEY}
• Two LLM calls per request:
  1. Reason about user intent
  2. Select and execute appropriate tool

🎓 Learning Outcomes

This project demonstrates:

• MCP protocol implementation and best practices
• WebSocket real-time communication
• LLM integration with fallback patterns
• Streaming/chunked responses
• AI agent reasoning patterns
• Async/await C# patterns
• Real-time command center UX

📝 License

MIT License - see LICENSE file for details

🙏 Acknowledgments

• Anthropic - MCP specification
• Google AI - Gemma 4 LLM API
• .NET Foundation - ASP.NET Core

Classroom demo for MCP protocol and AI agent orchestration 🎓