Post

HTB: DevHub

A HackTheBox exploitation chain demonstrating MCP server security vulnerabilities. The attack chain begins with an unauthenticated RCE endpoint on the MCPJam inspector service (port 6274) that spawns arbitrary processes from attacker-controlled commands. After gaining shell access, the attacker discovers a Jupyter Lab instance with its access token exposed in the systemd service file and process list, enabling lateral movement to a new user. Final privilege escalation involves accessing an internal MCP server running as root that exposes an undocumented tool capable of dumping the root user’s private SSH key. The write-up illustrates practical security risks in AI agent infrastructure, including credential exposure in service configurations and insufficient API authentication.

Read original article

This post is licensed under CC BY 4.0 by the author.