The MCP & AI Agent hackathon invited developers, tinkerers and open source enthusiasts from all backgrounds to get hands-on with building cool MCP servers and AI agents, exploring what is possible to make cloud native and our life more approachable and human centric through conversations using one or more open source projects: agentgateway, kagent, agentregistry and more. The hackathon ran two months, from early February to early April 2026. After carefully reviewing all of the submissions, we are excited to announce our winners and runner-ups. In many categories, the competition was extremely close, so we selected standout runner-ups alongside the winners.
Hackathon Winners 🥳🏆
Building Cool Agents track
This is my personal favorite track and huge congrats to Salvador Arreola (Mexico) for winning this category. He built Frugalia, an Agentic FinOps platform that uses kagent, kgateway and kmcp to autonomously detect, analyze, and resolve infrastructure waste in Kubernetes clusters. I enjoyed reading his blog, watching his project video and exploring the GitHub repository. This is a practical and impactful system that can continuously help teams optimize cloud resource spending in Kubernetes environments.
The runner-up goes to Oswaldo Gomez (Poland). He built a Kubernetes-native clinical AI agent platform that federates multiple MCP servers into a single endpoint, enabling IDEs and AI clients to securely query MIMIC-IV data through agents, tools, and service discovery. Check out his project video and GitHub repository to learn more.
Explore agentregistry track
The winner of this track is Denys Vasyliev (UK), a long-time contributor to kagent. He built a Kubernetes-native multi-cluster inventory controller for enterprise AI infrastructure, inspired by agentregistry. This project demonstrates strong thinking around scalable AI infrastructure management in cloud-native environments. Check out his project video and GitHub repository to learn more.
MCP & AI Agents Starter track
We created this track to encourage newcomers to MCP and AI agents to learn and build something impactful.
Huge congrats to John Capobianco (Canada) for winning this category. His team built a fun and creative agent with vision on top of kagent. His agent can access webcam and capture photos and transform them with Nano Banana and Veo3 into AI augmented media. A huge bonus - it even supports American Sign Language Mode (ASL) for accessibility. His demo video definitely made me smile – check out his GitHub repository to learn more.
The runner-up goes to Massimo Crippa (Italy), who built kagent-dotnet, a project that integrates kagent with the Microsoft agent framework using A2A, enabling developers to bring Microsoft Agent Framework–based agents into the kagent ecosystem. I enjoyed reading his blog, watching the project video and exploring the GitHub repository.
Open Source Contributions track
The winner of this track goes to Kevin Cao, Yitaek Hwang, and Jaden Lee for their outstanding contributions to agentgateway. They submitted 9 substantial PRs during their work to turn agentgateway from a model proxy into an identity-aware control plane for agents. Their contributions span identity at the edge, improved observable agent execution with open telemetry, and enhanced gateway to support multiple model providers.
Secure & Govern MCP track
The winner of this track goes to Huzefa Hamdard (India), who built an AI-Powered Kubernetes-native governance for MCP (Model Context Protocol) infrastructure. The project monitors agentgateway and kagent resources, evaluates security posture with an MCP-Server-centric scoring model, and provides AI-powered risk analysis via Google Gemini or Ollama. Check out the project video and Github repository to learn more.
The runner-up goes to Mehmet Hilmi Emel (Turkey), who built MCP.STORE, a full-stack e-commerce platform that demonstrates MCP authentication and authorization using agentgateway, Keycloak JWT, and CEL-based rules - enforced at the protocol level so AI agents cannot bypass security boundaries. I enjoyed reading his blog, watching his video and exploring his GitHub repository.
Acknowledgement
It was incredibly inspiring to review all the submissions. Many of these projects pushed creative boundaries and demonstrated real-world value in cloud-native AI agent systems.
A big thank you to all the judges for dedicating their time and effort to reviewing and scoring submissions thoughtfully. Your insights were invaluable!
Thank you to everyone who participated in the hackathon and shared your work with the community.
Special thanks to Rose Sawvel for her outstanding support in organizing and running the hackathon, and to my employer Solo.io for sponsoring and enabling this event.
%20a%20Bad%20Idea.png)
%20For%20More%20Dependable%20Humans.png)
