When you build an MCP server, you don’t control the agent calling it or the model powering it. The same server might be called by Claude Code, Cursor, or something completely custom, and it could be powered by anything from a small local LLM to a large foundation model.
In this episode of Datadog Illuminated, Reilly Wood, Staff Engineer at Datadog, explains how the team built the Datadog MCP server to work reliably across that entire wildcard space.
Every design decision flows from one constraint. You never know which agent/LLM combo will be calling the MCP server, so it has to work for all of them. That shapes everything from data formats (YAML carries the same data as JSON in far fewer tokens), to how many tools you expose, to what your error messages say.
This is the lightboard from my Datadog Illuminated episode with Julien Le Dem on OpenLineage.
Before OpenLineage, if your data broke somewhere in the pipeline, you had no way to trace it back to where it went wrong. OpenLineage fixes that by giving every tool a shared language for where data came from, so you can find the exact source of the problem instead of guessing.
“Managing Kubernetes RBAC manually is a path to burnout.”
In this episode of 🌩️Thunder, Saim Safdar, CNCF Ambassador and host of the Cloud Native Podcast, explains how Paralus centralizes Kubernetes access management across your clusters.
Paralus replaces per-cluster role juggling with a single layer for SSO, group-based access control, and full kubectl audit logs, whether you have one cluster or dozens across cloud providers and on-prem.
We secure data at rest. We secure data in transit. But what about data in use?
In this episode of 🌩️Thunder, Tobin Feldman-Fitzthum explains how Confidential Containers uses hardware-level trusted execution environments to isolate your Kubernetes workloads from everything outside them, including the Kubernetes control plane and the cloud provider itself.
Will AI help or hinder a developer’s experience with an internal platform? At KubeCon EU, the audience found out. My bestie Viktor Farcic (Upbound) and I ran an interactive session where attendees voted at key moments on which tools our AI agent would use. I built the demo to work with any combination they chose, on the fly.
The agent investigated a broken app, used semantic search to navigate cluster resources, and deployed on a developer’s behalf, with no cluster access required. Expect surprises, stumbles, and a real conversation about what AI is (and isn’t) ready for in platform engineering.
