I built this over the past few weeks for a cross-company CTF I was participating in. It was mostly an experiment to learn about CTFs, MCP servers, Kali Linux, Claude Code, and really just how far LLMs can go in a given domain.
It's basically just an MCP server (for integrating with the CTF server APIs, providing notes, and a few other niceties) paired with a Kali Linux-based Docker image that has Claude Code installed, plus a custom slash command [1] to tie it all together.
It performed admirably during the CTF I tried it on, it was able to zero-shot solve maybe 10 or so of the simpler challenges, and provided substantial assistance on another 5 or 6 before getting stuck. It didn't stand a chance against the hardest challenges.
This was the first CTF I've participated in, and it was an absolute blast. I can imagine some people feeling that LLMs take the fun out of CTFs, but I think the "centaur" [2] aspect of human-LLM interactions is both powerful and effective, given the right infrastructure and UX.
Happy to answer any questions people have about the project!
[1] https://docs.anthropic.com/en/docs/claude-code/slash-command...