VulnClaw

Why we included this project

Security teams tired of driving every step of a pentest by hand can give VulnClaw a plain-language request like 'run a pentest on this target' and let the CLI agent carry the engagement through recon, vulnerability discovery, PoC verification, and a structured report with a Python PoC script. The model decides which MCP tools to call next rather than following a fixed script, and raw tool output is kept in an evidence store, so a claimed finding only counts if it matches real results, a useful guard against hallucinated flags. Built-in skills cover CTF, web, intranet, and red-team scenarios, and it runs with OpenAI-compatible providers, including local Ollama. It is meant for authorized testing, CTF, and training, and the Python execution path is explicitly experimental rather than a hardened sandbox.