agent-opfor
Open-source adversary emulation tool that lets security teams and AI developers red-team AI agents and MCP servers via CLI, IDE, or browser extension to uncover prompt injection and jailbreak flaws.
Tool overview
agent-opfor is an open-source adversary emulation tool targeting AI agents, LLM applications, and MCP servers. It enables red teams and security engineers to mimic real attacker behaviors from the command line, inside their IDE, or through a browser extension, uncovering vulnerabilities such as prompt injection and jailbreaks. Built around OWASP and related guidelines, it covers common AI security test scenarios.
The tool offers multiple accessible interfaces: a CLI for automation pipelines, an IDE plugin for shift-left security during development, and a browser extension that lowers the barrier for non-experts on the team to participate in preliminary security checks. Licensed under Apache-2.0, users can freely audit, customize, and deploy it without licensing costs.
However, the project is at an early stage (8 stars, 2 forks) and comes with limited documentation and community support. It does not replace comprehensive security audits; its focus is on adversarial prompt simulation and may miss other AI risk domains. Misuse of the tool could also disrupt internal systems, so operational guardrails are advisable.