AgentHound

Offensive security framework that maps and attacks AI agent infrastructure such as MCP servers, A2A endpoints, LiteLLM gateways, Ollama, Qdrant, MLflow and Jupyter. Performs discovery, credential and model looting, tool poisoning and MCP implants, then loads findings into a Neo4j graph for BloodHound-style attack-path analysis. Ships as a single static Go binary with mutating operations defaulting to dry-run.