Lab

Isovalent Enterprise: Connecting and Securing MCP Servers on Kubernetes

Observe and secure Model Context Protocol (MCP) servers on Kubernetes with Cilium, Hubble, and Tetragon.

The Model Context Protocol (MCP) is how AI agents reach tools and systems, and increasingly those MCP servers run on Kubernetes. This lab looks at MCP from a platform engineer's seat: how MCP servers talk, where they run, and how you observe and secure them.

You will run a Kubernetes MCP server over HTTP and watch its calls with Hubble, then restrict it with a Cilium NetworkPolicy. You will then run the same job over a local stdio server, discover why the network tools go blind, and bring in Tetragon at the process layer. Along the way you will meet a real access-control bypass (CVE-2026-46519) and, in the exam, write the Tetragon policy that stops it.

  • Introduction to MCP

    Meet the Model Context Protocol, and see why it exists

  • 🚦 MCP Transports and the Remote Server

    Learn stdio vs HTTP, then deploy the Kubernetes MCP server

  • 🛰️ Observe and Secure the MCP Server

    See MCP calls in Hubble, then restrict the server with a Cilium policy

  • 📚 Resources and Egress Control

    Meet the resources primitive, then limit an MCP server's egress with a Cilium Layer 7 policy

  • 🕳️ The Local Server and the Blind Spot

    Run MCP over stdio, lose network visibility, and meet a real CVE

  • 🏆 Final Exam: Stop the CVE

    Turn observation into enforcement with Tetragon