Metasploit for AI agents
HackMyAgent is the offensive security toolkit for AI agents. It scans an agent or MCP server, attacks it with real adversarial payloads, and fixes what it finds. Open source under Apache-2.0.
Why Metasploit
Metasploit gave security teams a single offensive toolkit to probe systems the way an attacker would. HackMyAgent does that for AI agents. A red-team run sends real adversarial payloads at a live agent and reports exactly what landed. Against the LegacyBot target in the Damn Vulnerable AI Agent, all 28 of 28 payloads in a passive run succeeded across 14 categories.
Scan
209 static checks plus 29 semantic checks find hardcoded credentials, exposed endpoints, MCP misconfigurations, and supply chain risk.
Attack
164 adversarial payloads run real prompt injection, jailbreak, and tool-abuse attacks against a live agent, then report what landed.
Fix
Every finding ships with a verify command and a runnable fix. Auto remediation can apply the fix with rollback.
A HackMyAgent scan reports a score, the critical findings, and the path to fix them.
Run it
Scan the current project, no flags required:
npx hackmyagent secureRun a red-team attack against a live agent endpoint:
npx hackmyagent attack http://localhost:7003 --intensity passive