AARI World Model + GDC Security Audit

Producer mindset. Code-first. No GUIs.

A physics-based infrastructure digital twin and security assessment toolkit for the Garage Data Center (GDC). Built for AARI students to learn real infrastructure engineering through AI-native tooling.

What This Is

This repository contains three MCP (Model Context Protocol) servers that turn Claude Code into an infrastructure operator:

| Server | What It Does | |--------|--------------| | Datacenter | Simulates thermal dynamics, detects anomalies, predicts failures | | Deployment | Deploys applications to OpenShift/Kubernetes clusters | | Security | Runs Kali Linux security tools (nmap, nikto) for GDC audits |

Why MCP instead of Streamlit?
Streamlit teaches students to click buttons. MCP teaches students to build AI-native automation. Same pattern used in production systems.

Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Test the Physics Engine

python -m src.physics

Expected output:

AARI World Model: Thermal Physics Demo
==================================================
[Normal] Peak temp: 65.0°C
[Failure] Peak temp: 178.9°C

3. Run with Claude Code

# Start Claude Code in this directory
claude

# Then ask it to:
> Run a thermal simulation with fan failure at step 30
> Analyze the telemetry for anomalies
> Deploy the digital twin to OpenShift

The Physics

The datacenter simulation models Newton's Law of Cooling with failure cascades:

dT/dt = (k_heating × load - k_cooling × (T - T_ambient) × efficiency) / thermal_mass

Key Parameters

| Parameter | Value | Description | |-----------|-------|-------------| | k_heating | 0.08 | Heat generated per % load per minute | | k_cooling | 0.15 | Cooling coefficient | | T_ambient | 22°C | Base ambient temperature | | critical_temp | 85°C | Emergency shutdown threshold | | fan_degradation | 0.02/min | Efficiency loss during failure |

Failure Cascade

When a fan fails at step 20:

1. Step 20: Fan efficiency starts degrading (1.0 → 0.98 → 0.96...)
2. Step 55: Temperature hits warning threshold (75°C)
3. Step 70: Temperature hits critical threshold (85°C)
4. Step 100: Runaway to 178°C without intervention

This is why predictive monitoring matters - catching the trend at step 30 gives you 40 minutes to respond.

MCP Server Details

Datacenter Server (mcp-servers/datacenter/server.py)

Tools:

• run_simulation - Execute thermal simulation with configurable parameters
• detect_anomaly - Multi-method anomaly detection (threshold, rate-of-change, statistical)
• full_analysis - Complete pipeline: simulate → detect → predict

Example via Claude Code:

> Run simulation with 150 steps, inject fan failure at step 30, then analyze for anomalies above 70°C

Deployment Server (mcp-servers/deployment/server.py)

Tools:

• check_login - Verify OpenShift cluster access
• get_pods - List pods in a project
• deploy_app - Deploy container image
• get_routes - Get public URLs
• apply_yaml - Apply Kubernetes manifests

Prerequisites:

# Install OpenShift CLI
brew install openshift-cli  # macOS
# or download from Red Hat

# Login to GDC
oc login https://api.gdc.local:6443 --token=<your-token>

Security Server (mcp-servers/security/server.py)

Tools:

• check_tools - Verify available scanners (nmap, nikto, gobuster)
• port_scan - TCP port scanning
• service_scan - Service version detection
• vuln_scan - Vulnerability assessment
• web_scan - Web application scanning
• generate_report - Compile audit summary

Prerequisites:

# On Kali Linux or install tools manually
apt install nmap nikto gobuster

⚠️ IMPORTANT: Only scan systems you own or have explicit permission to test.

Project Structure

aari-world-model/
├── src/
│   ├── __init__.py          # Package exports
│   ├── physics.py           # Thermal simulation engine
│   ├── generator.py         # Telemetry data pipeline
│   └── anomaly.py           # Anomaly detection algorithms
│
├── mcp-servers/
│   ├── datacenter/
│   │   └── server.py        # Simulation MCP server
│   ├── deployment/
│   │   └── server.py        # OpenShift MCP server
│   └── security/
│       └── server.py        # Kali tools MCP server
│
├── claude-tasks/
│   ├── analyze-thermal-anomaly.md
│   ├── deploy-to-openshift.md
│   └── security-audit-gdc.md
│
├── infra/
│   ├── Dockerfile           # Multi-stage container build
│   └── deployment.yaml      # OpenShift/K8s manifests
│
├── data/                    # Generated telemetry (gitignored)
├── reports/                 # Generated analysis reports
├── mcp-config.json          # Claude Code MCP configuration
└── requirements.txt         # Python dependencies

Claude Tasks

Pre-built task definitions in claude-tasks/ that Claude Code can execute autonomously:

analyze-thermal-anomaly.md

Simulates a failure scenario and generates an incident report.

deploy-to-openshift.md

Deploys the digital twin to the GDC OpenShift cluster.

security-audit-gdc.md

Runs a security assessment against GDC infrastructure.

Run a task:

claude "Execute the thermal anomaly analysis task"

Deploy to GDC

Build Container

docker build -t aari-world-model:v1 -f infra/Dockerfile .

Deploy to OpenShift

# Login to Dave's cluster
oc login https://api.gdc.local:6443

# Apply manifests
oc apply -f infra/deployment.yaml

# Verify
oc get pods -n aari-world-model
oc get route -n aari-world-model

Or via Claude Code

claude "Deploy the digital twin to OpenShift and verify it's running"

For AARI Students

This project teaches:

1. Physics First - Understand the math before touching tools
2. Data Pipelines - Generate, transform, and analyze telemetry
3. Anomaly Detection - Multiple methods, defense in depth
4. MCP Protocol - How AI agents invoke tools
5. Container Deployment - Docker, Kubernetes, OpenShift
6. Security Assessment - Ethical penetration testing

Learning Path

| Week | Focus | Files | |------|-------|-------| | 1 | Physics simulation | src/physics.py | | 2 | Data generation | src/generator.py | | 3 | Anomaly detection | src/anomaly.py | | 4 | MCP development | mcp-servers/*/server.py | | 5 | Container deployment | infra/ | | 6 | Security auditing | mcp-servers/security/ |

The AARI Way

"The cloud is just someone else's computer. Stop renting. Start building."

This project exists because:

• AWS won't teach you infrastructure
• Bootcamps won't teach you physics
• Universities won't teach you production

AARI teaches all three.

Links

• AARI Website: atlanta-robotics.org
• MCP Protocol: modelcontextprotocol.io
• OpenShift Docs: docs.openshift.com

License

Apache 2.0 - Build freely, deploy anywhere.

Questions? nolan@atlanta-robotics.org