What the MCP Hackathons Are Actually About

Solo.io's MCP & AI Agents hackathon produced eight winners, and the track structure alone tells you where attention sits: Building Cool Agents, agentregistry, Starter, OSS Contribution, and MCP Security & Governance. Nearly half the categories are about operations and control rather than what the agent can do.

The winners follow the same line. The MCP Security & Governance winner sits on the monitoring side of agent-to-server traffic; the agentregistry winner handles taking inventory of agents. Frugalia has an agent find and fix wasted Kubernetes spend, but it too is infrastructure work, built by combining kagent, kgateway and kmcp. And as SprintForge showed by wiring together the Microsoft Learn and Jira MCP servers, connecting multiple servers is itself the technical highlight now.

Takeaway: in this space, designing who governs the agent lands harder than designing what the agent can do. Authentication, least privilege, audit trails, an inventory of what is running — unglamorous, but exactly where companies get stuck, and exactly what the platform engineers judging you value most.