VulnClaw
An AI penetration-testing agent that turns a natural-language request into an automated recon, vulnerability discovery, exploitation and report-generation workflow, for authorized testing and CTF use.
VulnClaw is a standalone AI penetration-testing agent built on an LLM agent plus an MCP tool chain and optional skill references. Given a natural-language instruction, it runs through information gathering (fingerprinting, port scanning, directory enumeration), vulnerability discovery (injection points, known CVEs, misconfigurations), exploitation with proof-of-concept verification, and structured report generation with an accompanying proof-of-concept script. It is intended for authorized penetration testing, CTF competitions, security education and red-team exercises.
It supports fourteen LLM providers, including OpenAI, Anthropic, DeepSeek and a local Ollama option, and ships four MCP services, two of them (fetch and memory) built in and two (chrome-devtools and burp) connecting to external services for browser automation and HTTP traffic replay. Findings are checked against raw tool output before being accepted, so a claimed flag or conclusion must appear verbatim in the evidence, and the project includes fifty specialized reference skills covering CTF, web, internal-network, reverse-engineering and authorized red-team topics.
It is written in Python and released under the MIT license.