AI-Tooling

Red-team tooling and reference material for AI systems.

MCP servers

Original MCP servers, written from scratch against each tool's official API/driver:

• mythic-mcp/: operator actions against a Mythic C2 server
• bloodhound-mcp/: attack-path queries against a BloodHound Neo4j database
• nmap-mcp/: host discovery and port/service scans via nmap, with a scope guard

Each is an independent Python project with its own pyproject.toml, README, and .env.example.

Reference material

• prompt-injection/: a taxonomy of system-prompt / secret-extraction techniques (OWASP LLM07), a payload list, and a deliberately-vulnerable practice app with a runner that reports which payloads leaked its canary secret.
• insecure-output/: XSS and client-side bugs from LLM output rendered into HTML without encoding (OWASP LLM02). Payloads, worked reflected and stored scenarios, and a practice app (LLMShip) with a built-in exfil listener and a probe script.

Authorized use only

Everything here is for lab environments, CTFs, and engagements where you already have permission to operate against the targets and instances involved.