In this Office Hours session, we explored the evolution of generative AI, the emergence of agentic systems, and the game-changing impact of Model Context Protocol (MCP) on DevOps. We discussed real-world demos of how Claude (LLM) can now interface with your terminal, Podman, and Kubernetes to automate real tasks—proving that Agentic DevOps is no longer theoretical, it’s here.
What if your favorite LLM could not only talk smart but act smart?
That’s what we explored in our latest Office Hours session. We dove deep into how the AI world is transitioning from just large language models (LLMs) to agentic systems—LLMs that don’t just talk but also do. And at the center of this revolution? A new standard called Model Context Protocol (MCP).
🧠 From LLMs to Agentic AI – Giving Intelligence a Body
In 2023, we all saw the power of LLMs explode. Then came the RAG wave—Retrieval-Augmented Generation, giving LLMs external context to answer organization-specific questions.
But RAG alone can’t act.
Enter 2024-2025: the age of Agentic AI, where we don't just ask questions—we get stuff done. Agentic AI gives our LLMs hands and legs—like a human who can now take action in the real world.
Think of your LLM like a brainy assistant. But until now, it couldn’t really do much. That changes with MCP.
🔌 MCP – The USB-C of the Agentic World
What if instead of coding integrations for each new tool you want your agent to interact with, you could just plug it in like a USB device?
That’s MCP. Created by Anthropic (makers of Claude), MCP is an open standard that allows tools to expose their capabilities to LLMs in a structured way. It’s like the USB-C port of AI.
Want to connect to your terminal? Use the terminal MCP server.
Want Podman access? Use the Podman MCP server.
Kubernetes? Git? Monitoring tools? There are MCP servers for all of them—and counting.
In fact, we’ve already curated a list of MCP servers just for DevOps tools, and the ecosystem is growing daily.
🛠 Live Demos: From Talking to Doing
We went hands-on with Claude’s MCP-enabled desktop app and demonstrated:
Connecting Claude to the Mac terminal using an MCP server.
Asking it to list top memory-hogging processes (it wrote and executed the shell command on its own).
Spinning up and tearing down a Podman-based NGINX container on demand.
Building a multi-node Kubernetes cluster using
kindbased on custom instructions in a PDF.Running full workflows via Claude—from cloning repos, reading YAML config, setting up clusters, to validating nodes—all autonomously.
Debugging a crashing Kubernetes pod by reading logs, identifying a missing env variable, and auto-fixing it!
All of this was possible just by installing MCP servers and configuring Claude with them.
🔄 Replacing AIOps with Open Agentic DevOps
We’re entering a world where open-source agentic systems can replace vendor-locked AIOps tools. With standards like MCP:
You can build your own DevOps agents.
Connect them to open-source tools like Kubernetes, Prometheus, or GitHub.
Provide natural language instructions + context (RAG-style), and let your AI agent handle the rest.
This means we’re finally seeing the democratization of AIOps and the birth of open Agentic DevOps.
🔐 Guardrails, Security & Control
Yes, these agents are powerful—but you can and should define guardrails:
Restrict or allow specific actions.
Define capability scopes in your MCP server.
Add approval workflows where needed.
This makes MCP not just powerful but safe and enterprise-ready.
🌱 So What Now?
We’re at the very beginning of this revolution. The MCP ecosystem is raw, evolving fast, and filled with potential. And if you're a DevOps engineer, this is your chance to be early in what will soon be the default paradigm for automation.
💡 Whether you want to build your own MCP server, an agentic client, or a DevOps co-pilot — now is the time.
Coming Soon
I'm building a full course on MCP and Agentic DevOps at SchoolOfDevops.com where you’ll learn how to build, configure, and extend your own agentic systems step by step. Stay tuned!
If you liked this update and want more such breakdowns, subscribe to this Substack for deep dives, tutorials, and real-world DevOps + AI content.
✌️
Gourav Shah
Founder, School of DevOps & AI
Helping you thrive in the AI era
Share this post