Zero-Trust IP Allowlisting — All admin operations (scans, audits, infra controls) are locked behind IP allowlisting. Even with valid credentials, requests must originate from the home network. VPN, proxy, and foreign IPs are rejected. IPv6 prefix matching requires a minimum /64 specificity to prevent broad-prefix bypass.

SIC — Security Intelligence Center

AI-Powered Pentesting MCP Framework for Authorized Security Testing

An AI-powered pentesting MCP framework with 150+ security tools and 12+ autonomous agents for authorized security testing, CTF challenges, and defensive research.

SIC Engine | Installation | API Reference

---

Overview

SIC is an AI-powered penetration testing framework that runs as a local server, exposing a comprehensive API and MCP interface for integration with AI clients (Claude, GPT, Copilot, Cursor, etc.).

---

Architecture

SIC runs 150+ real offensive security tools (nmap, sqlmap, nuclei, hydra, etc.) — fully sandboxed in a hardened Docker container with multiple security layers.

How It Works

AI Client (Claude, GPT, Copilot, Cursor)
  │
  ▼ (MCP Protocol)
SIC MCP Server
  │  ├─ Intelligent decision engine
  │  ├─ Tool selection & parameter optimization
  │  └─ Attack chain discovery
  │
  ▼
127.0.0.1:9888 (loopback only — never exposed)
  │
  ▼
Docker Container (sic-scanner)
  │  ├─ Scope enforcer (ALLOWED_TARGETS whitelist)
  │  ├─ Dry-run gate (on by default)
  │  └─ Tool execution (150+ tools)
  │
  ▼
./output/ (results only — source baked into image)

Container Isolation

The Docker container enforces 12 security controls:

Control	Setting	Purpose
Port binding	127.0.0.1:9888	Never reachable from network
User	scanner (uid 1001)	Non-root, no privilege escalation
Capabilities	cap_drop: ALL	Zero Linux capabilities
Privilege escalation	no-new-privileges: true	Blocks setuid/setgid
CPU limit	2 cores	Prevents self-DoS
Memory limit	2 GB	Bounded resource usage
DNS	127.0.0.1 only	Blocks external hostname resolution
Network	scanner-net bridge (internal on Linux)	No cross-container routes
Scanner mode	SCANNER_MODE=sandbox	Restricts target scope at app layer
Allowed targets	target.example.com,192.168.1.0/24	Whitelist-only scanning
Request budget	MAX_REQUESTS_PER_SCAN=500	Prevents runaway scans
Dry-run default	DRY_RUN_DEFAULT=true	Must explicitly opt into live scans
Scan timeout	300s hard wall	Kills scans after 5 minutes
Volume mounts	./output only	Source code baked into image, never mounted

Multi-Stage Build

The Dockerfile uses 3 stages to keep the image lean and the build fast:

Stage	Base	What It Builds
go-builder	golang:1.24-alpine	13 Go tools (ffuf, gobuster, nuclei, httpx, subfinder, katana, etc.)
py-builder	python:3.12-slim	30+ Python packages (sqlmap, dirsearch, theHarvester, pwntools, etc.)
runtime	python:3.12-slim	Final image — all tools + HexStrike API server

Heavy packages (angr, autorecon, spiderfoot) are stubbed — the System Tab runs which <tool> to show availability, so stubs satisfy that without the OOM risk.

Running It

# Start the sandboxed container
cd docker/sic-scanner
docker compose up -d

# Verify health
curl http://127.0.0.1:9888/health

# Register with PM2 (optional)
pm2 start "docker compose -f docker/sic-scanner/docker-compose.yml up" --name sic-scanner

---

SIC Engine

AI-powered penetration testing framework with MCP protocol support. Connects to Claude, GPT, Copilot, Cursor, or any MCP-compatible AI client.

Architecture

graph TD
    A[AI Agent - Claude/GPT/Copilot] -->|MCP Protocol| B[SIC MCP Server v6.0]

    B --> C[Intelligent Decision Engine]
    B --> D[12+ Autonomous AI Agents]
    B --> E[Modern Visual Engine]

    C --> F[Tool Selection AI]
    C --> G[Parameter Optimization]
    C --> H[Attack Chain Discovery]

    D --> I[BugBounty Agent]
    D --> J[CTF Solver Agent]
    D --> K[CVE Intelligence Agent]
    D --> L[Exploit Generator Agent]

    B --> P[150+ Security Tools]
    P --> Q[Network - 25+]
    P --> R[Web App - 40+]
    P --> S[Cloud - 20+]
    P --> T[Binary - 25+]
    P --> U[CTF - 20+]
    P --> V[OSINT - 20+]

Loading

How It Works

1. AI client sends commands via MCP protocol
2. Decision engine selects optimal tools and parameters
3. Security tools execute scans, exploits, and analysis
4. Results formatted and returned through MCP with visual output

AI Agents

Agent	Capability
BugBounty Agent	Automated bug bounty hunting workflow
CTF Solver Agent	Challenge analysis and solution strategies
CVE Intelligence Agent	CVE lookup, exploitability analysis, patch tracking
Exploit Generator Agent	Proof-of-concept exploit development
Recon Agent	Automated reconnaissance and asset discovery
Web Scanner Agent	Comprehensive web application assessment
Cloud Auditor Agent	Multi-cloud security posture review
Network Agent	Internal/external network penetration testing
Forensics Agent	Digital forensics and incident response
OSINT Agent	Open-source intelligence gathering
Social Engineering Agent	Phishing simulation and awareness
Report Generator Agent	Automated pentest report creation

Security Tools Arsenal

Network Security (25+ tools)

nmap, masscan, rustscan, netcat, tcpdump, wireshark-cli, arp-scan, ping sweep, traceroute, DNS zone transfer, subdomain enumeration, and more.

Web Application Security (40+ tools)

sqlmap, nikto, wfuzz, gobuster, feroxbuster, httpx, nuclei, XSS detection, SSRF scanner, CORS checker, directory brute-forcing, and more.

Cloud Security (20+ tools)

ScoutSuite, Prowler, CloudSploit, S3 bucket scanner, IAM analyzer, container security scanning, and more.

Binary Analysis (25+ tools)

GDB, Radare2, Ghidra, Binwalk, checksec, ROPgadget, pwntools, and more.

CTF Tools (20+ tools)

CyberChef, John the Ripper, Hashcat, Stegsolve, memory/disk forensics toolkit, and more.

OSINT (20+ tools)

theHarvester, Shodan, SpiderFoot, Recon-ng, Maltego, and more.

Advanced Capabilities

• Intelligent Decision Engine — AI-driven tool selection based on target context
• Parameter Optimization — Automatic tuning per tool/target combination
• Attack Chain Discovery — Links vulnerabilities into exploitable chains
• Smart Caching — Avoids redundant scans, caches intermediate results
• Resource Management — CPU/memory-aware scheduling
• Error Recovery — Automatic retry with fallback strategies

---

Installation

SIC Engine

git clone https://github.com/DevCraftXCoder/sic.git
cd sic
pip install -r requirements.txt
python sic_launcher.py

Default: http://127.0.0.1:5000

# Verify
curl http://127.0.0.1:5000/health

# Debug mode
python sic_launcher.py --debug

# Custom port
python sic_launcher.py --port 5001

# Register with PM2
pm2 start sic_launcher.py --name sic-server --interpreter python

PowerShell Install

Load the helper functions in any PowerShell session by dot-sourcing the bundled profile:

. "C:\path\to\sic\install\sic-profile.ps1"

To load automatically on every PowerShell session, open your profile file and add the line permanently:

notepad $PROFILE

Then add this line inside the file (adjust the path to match your clone location):

. "C:\Za\sic\install\sic-profile.ps1"

Helper Functions

Function	Usage	Description
sic-scan	sic-scan -Target 192.168.1.1 [-Type web]	Submit a smart scan via /api/intelligence/smart-scan
sic-health	sic-health	GET /health — full server health and telemetry
sic-incidents	sic-incidents	GET /api/incidents — list open incidents
sic-fix	sic-fix <finding-id>	POST /api/command — AI-assisted remediation for a finding
sic-version	sic-version	Show server version and status (from /health)

All helpers target http://127.0.0.1:9888 by default (the SIC_PORT default). Override by setting $SIC_BASE before dot-sourcing.

AI Client Integration

Claude Desktop / Cursor:

{
  "mcpServers": {
    "sic": {
      "command": "python",
      "args": ["/path/to/sic/sic_mcp.py"]
    }
  }
}

VS Code Copilot:

{
  "mcp.servers": {
    "sic": {
      "command": "python",
      "args": ["/path/to/sic/sic_mcp.py"]
    }
  }
}

---

API Reference

SIC Engine Endpoints

Endpoint	Method	Description
/health	GET	Full system health + telemetry
/api/tools	GET	List all available tools
/api/tools/<name>	GET	Tool detail and status
/api/scan	POST	Run a targeted scan
/api/agents	GET	List AI agents
/api/agents/<name>/run	POST	Execute an agent task
/api/processes	GET	List running processes
/api/processes/<id>	DELETE	Kill a process
/api/cache/clear	POST	Clear scan cache

Common MCP Tools

# Network
sic_nmap_scan(target, flags)
sic_masscan(target, ports)
sic_port_scan(target)

# Web
sic_nikto_scan(target)
sic_sqlmap(target, params)
sic_directory_bruteforce(target, wordlist)

# Recon
sic_subdomain_enum(domain)
sic_whois(domain)
sic_dns_lookup(domain)

# Vulnerability
sic_nuclei_scan(target, templates)
sic_cve_lookup(cve_id)
sic_exploit_search(query)

---

Security & Responsible Use

SIC is an offensive security toolkit for professionals. It generates real exploits, runs real scans, and can cause real damage if misused.

• All tools run locally — no telemetry, no data exfiltration
• API key auth protects the server
• Smart caching stores results locally — clear with /api/cache/clear
• Exploit generation and CVE research are first-class features

This is a private tool shared among trusted peers. If you have access, you already know the rules: test only what you're authorized to test.

---

Recent Additions

• Security hardening (v6.0.0) — 38 tool routes patched for shell injection safety; zero injection vectors remain across all endpoints
• Per-service API tokens — Individual tokens per integrated service with expiry warnings (#25 token-expiry system)
• Rate limiting — All endpoints rate-limited; 25 CVEs addressed via pip-audit dependency audit
• Admin panic-stop — Admin role gate on /api/admin/panic-stop (was unauthenticated)
• Generic error responses — P0 scope_status and internal error details no longer leaked to API consumers
• Auth/tier/scope enforcement — P1 complete: request validation wired across all tool routes
• Claude Code /sic command — Direct Claude Code integration for interactive pentesting sessions (#27)
• Bugs archive endpoint — GET /api/bugs-archive for historical bug/finding retrieval (#10)
• Zero-Trust IP Allowlisting — All admin operations locked behind IP allowlisting; IPv6 prefix matching requires minimum /64 specificity

---

License

MIT License — see LICENSE for details.