使用说明
🤖 MCP-CTF-Demo
I Brought Kali. They Brought AI.
How MCP Agents Are Rewriting CTFs — and What Banks Must Learn Before Attackers Do
Live Demo: siteq8.github.io/MCP-CTF-Demo
💀 The Story
At a recent CTF competition, I arrived prepared — Kali Linux loaded, custom exploit scripts ready. Then I watched competing teams deploy MCP-connected AI agents that chained reconnaissance, vulnerability discovery, and exploitation at machine speed, capturing flags at 3-5× my rate.
That moment revealed: the offensive security playbook has been rewritten.
🔌 What's Inside
MCP Security Server (scripts/mcp-server/)
A Python MCP server exposing 6 security tools:
| Tool | Description |
|---|---|
recon_dns |
DNS enumeration — subdomains, MX, TXT, NS records |
recon_ports |
TCP port scan — top 22 ports, banner grabbing |
recon_headers |
HTTP security header analysis (OWASP) |
vuln_ssl |
SSL/TLS configuration assessment |
exploit_chain |
Attack path analysis with MITRE ATT&CK mapping |
report_generate |
Structured findings report |
Interactive Demo GUI (docs/)
6-tab web dashboard for the SANS presentation:
- 📊 Overview — The CTF story + MCP explanation
- 🔌 MCP Architecture — Protocol diagram + tool reference
- 🎯 Live Demo — Simulated AI agent running all 6 tools
- ⚡ Human vs AI — Side-by-side speed comparison
- 🔗 Attack Chains — 3 multi-step attack paths with MITRE mapping
- 🛡️ Bank Playbook — 6-point defense strategy
🚀 Quick Start
Run the MCP Server
git clone https://github.com/SiteQ8/MCP-CTF-Demo.git
cd MCP-CTF-Demo
python3 scripts/mcp-server/mcp_security_server.py --port 8080
Call a Tool
curl -X POST http://localhost:8080 \
-H "Content-Type: application/json" \
-d '{"method":"tools/call","params":{"name":"recon_dns","arguments":{"domain":"example.com"}}}'
List Available Tools
curl -X POST http://localhost:8080 \
-H "Content-Type: application/json" \
-d '{"method":"tools/list"}'
🛡️ 6-Point Banking Defense Playbook
- Red Team WITH AI Agents — Deploy MCP agent pentesters in CI/CD
- Assume AI Speed — Automated containment within 30 seconds
- Behavioral Detection — UEBA over signatures
- Secure Your AI Agents — Zero-trust for MCP servers
- Update Threat Models — Add AI-augmented adversary scenarios
- Internal CTFs — Monthly AI-augmented red team exercises
⚡ Human vs AI Agent
| Task | Human Pentester | MCP AI Agent |
|---|---|---|
| DNS Recon | 15-30 min | 0.5 sec |
| Port Scan | 10-45 min | 2 sec |
| Header Analysis | 5-10 min | 0.3 sec |
| SSL Check | 5-10 min | 0.5 sec |
| Attack Chain Analysis | 1-2 hours | 1 sec |
| Report Writing | 2-4 hours | 2 sec |
| Total | 4-8 hours | ~8 seconds |
👤 Author
Ali AlEnezi · @SiteQ8 · 3li.info
Security Architecture Principal · National Bank of Kuwait (NBK Group) 🇰🇼
GPEN · GWEB · GDSA · GICSP · GCCC · CMU CISO · PCI DSS Professional
