The value is the loop rather than the toolbox: the model picks the nmap flags, reads the structured output that comes back, and decides whether the next step is a directory fuzz or an exploit search — instead of telling you which command to type. The container is what makes that workable on a laptop that is not running Kali, and it is also the boundary worth holding in mind, because every one of these tools acts on whatever target gets named.
A server that wraps the usual Kali Linux command line — whois, dnsenum, amass, whatweb, netdiscover, nmap, enum4linux, nikto, gobuster, dirb, ffuf, wfuzz, sqlmap, wpscan, hydra, john, hashcat, searchsploit, tcpdump and the Metasploit console — as tools with named parameters, returning each run's standard output and error to the model. The published image is built on kalilinux/kali-rolling with the toolchain installed and the rockyou wordlist already unpacked, so it behaves the same on Windows and macOS as on Linux with none of those binaries installed locally. Run directly on a host instead and it expects them on the PATH; packet capture needs elevated privileges either way. Long-running tools have their own timeouts, so an unresponsive target returns an error rather than hanging the conversation.
- Reconnaissance: whois records, DNS enumeration, passive subdomain discovery, web technology and header fingerprinting, and live hosts on a range
- Port and service scanning with the nmap flags you pass through, plus Windows and Samba enumeration and SMB share listing
- Web testing: vulnerability scanning, directory and content brute forcing, parameter fuzzing, SQL injection detection, and WordPress enumeration
- Password work: login brute forcing against a named service, and hash cracking with either john or hashcat against a wordlist you supply
- Exploit-DB search, packet capture on an interface, and a Metasploit module run with its options passed as key-value pairs
One command — docker pull cyberillo/kali-mcp-server:latest