video1.0<iframe src="https://www.loom.com/embed/9cb5cb6d1e604490a0f589e2820cf58c" frameborder="0" width="1920" height="1440" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>14401920Loomhttps://www.loom.com14401920https://cdn.loom.com/sessions/thumbnails/9cb5cb6d1e604490a0f589e2820cf58c-fdd7860283475c87.gif182.983In this video, I discuss the development of Flint, a session firewall designed for MCP-connected agents. Current controls fail to address risks that arise from the sequence of authorized actions, as demonstrated in incidents like the Superbase and Cursor attacks. Flint implements a two-layer system: a role-based access control layer and a behavioral detection layer that tracks data flow across tool calls. I have already built the role-based detection and several behavioral detection rules, and I am currently working on an inline gateway to enhance security. I encourage you to provide feedback on these developments as we move forward.