k8s-mcp-server

MCP.Pizza Chef: alexei-led

Eight tools ship inside a Docker image: four explain what a command does, and four actually run kubectl, Helm, istioctl, or argocd against whichever cluster your existing settings point at. That includes real changes — scaling, deploying, deleting — not just looking, so the read-only mount shown in the setup instructions protects your credentials file, not your cluster. No key or account is required; it reuses access you already have. Docker is needed, and clusters hosted on Amazon, Google, or Azure take extra setup. Last updated February 2026.

Coding
Other

Use This MCP server To

Ask why a pod keeps restarting and read its logs List what is running in a namespace without typing commands Install or upgrade a Helm chart by describing it Scale a deployment up or down from chat Get a plain-English explanation of a confusing cluster error Check which clusters my current settings can reach

README

K8s MCP Server

CI Status Release Status codecov Image Tags Image Size Python Version License: MIT

K8s MCP Server is a Docker-based server implementing Anthropic's Model Context Protocol (MCP) that enables Claude to run Kubernetes CLI tools (kubectl, istioctl, helm, argocd) in a secure, containerized environment.

Demo: Deploy and Troubleshoot WordPress

Session 1: Using k8s-mcp-server and Helm CLI to deploy a WordPress application in the claude-demo namespace, then intentionally breaking it by scaling the MariaDB StatefulSet to zero.

Session 2: Troubleshooting session where we use k8s-mcp-server to diagnose the broken WordPress site through kubectl commands, identify the missing database issue, and fix it by scaling up the StatefulSet and configuring ingress access..

k8s-mcp-server FAQ

Do I need a key or a paid account?
No. It reuses the cluster settings already on your computer, so whatever access you have, it has.
Which apps does it work in?
Claude Desktop is the documented path, with a settings block you paste in. It can also run in a web-served mode for other clients.
Can I use this to troubleshoot a deployment that will not start?
Yes — this is its best use. It can pull the status, read logs, describe the failing object, and then explain in plain terms what went wrong.
Can it break my cluster?
Yes. Four of the eight tools run real commands, including ones that scale, deploy, and delete. Approve each command individually and try it on a test cluster first.
How hard is setup?
You need Docker installed and running, then paste one block into Claude Desktop and restart it. No building from source.
Does it work with clusters on Amazon, Google, or Azure?
Yes, but each needs extra configuration beyond the basic setup, which the project documents separately.
Is it still maintained?
Yes. It was updated in February 2026 and has automated testing in place.