Kubernetes MCP server
For teams that run their own Kubernetes cluster. Through Flowplane an AI assistant can list namespaces and the pods, deployments, statefulsets, services, ingresses, jobs, nodes and events in them, look at one object, read a pod’s log and check whether a rollout has finished. It can scale a deployment or statefulset and restart a rollout; deleting a pod waits for your approval by default.
- Tools
- 5 read · 3 change
- Sign-in
- One shared key
- Approval, by default
- 1 of 3 changes ask first
- MCP server
- Flowplane’s tools
What your assistant can do in Kubernetes
5 tools that read and 3 that change something in Kubernetes. Reads are allowed; a change marked “asks first” waits for the person’s approval by default, and your workspace’s rules decide the rest.
Reads5
- List namespacesThe cluster's namespaces
list_namespaces - List workloads and morePods, deployments, services, jobs, nodes, events
list_resources - Look at one objectA summary, or the whole object
get_resource - Read a pod's logThe last lines of a container's log
get_logs - Check a rolloutWhether a deployment has finished rolling out
rollout_status
Changes3
- Scale a workloadSet how many replicas a workload runs
scale - Restart a rolloutReplace its pods one by one
restart_rollout - Delete a podasks firstIts controller usually makes a new one
delete_pod
Sign-in
The cluster is reached with the token of a ServiceAccount you make, bound to a Role with only the rights you want assistants to have. The workspace pastes it once; the assistant reads and changes only what that account may. Flowplane keeps it sealed; the assistant never sees it.
In Flowplane: In the Flowplane console, add Kubernetes, type the API server’s address (the server line of your kubeconfig), paste the ServiceAccount token and, if the cluster has its own certificate authority, its CA certificate. A cluster behind a firewall connects through a tunnel.
Kubernetes and MCP
The Kubernetes project has no official MCP server. There are community-built ones, such as containers/kubernetes-mcp-server, which you run yourself with your own credentials. Flowplane talks to your cluster’s API server directly, so there is nothing to install in the cluster.
Things to ask
- “Which pods in the production namespace are not running?”
- “Show the last 100 lines of the api pod’s log.”
- “Has the rollout of the web deployment finished?”
- “Scale the worker deployment in staging to 3 replicas.”
- “What warning events were there in the payments namespace?”
Connect your assistant to Kubernetes
One Flowplane address works in every assistant below, with the same rules and the same log. Step-by-step setup for each:
- Connect Claude to KubernetesClaude on the web and in the desktop app→
- Connect Claude Code to KubernetesAnthropic’s coding agent in the terminal→
- Connect ChatGPT to KubernetesChatGPT on the web→
- Connect Cursor to KubernetesThe Cursor code editor→
- Connect Codex to KubernetesOpenAI’s coding agent, in the terminal and the IDE→
- Connect GitHub Copilot to KubernetesCopilot Chat in VS Code, in agent mode→
- Connect Mistral Le Chat to KubernetesMistral’s assistant (Le Chat, now called Vibe)→
Questions
- Can the assistant read my Secrets?
- No. A Secret is listed by name, type and age only; its data is never read into an answer. A ConfigMap shows the names of its keys, not their values. To keep Secrets from being listed at all, give the ServiceAccount no right on them.
- What can it change in the cluster?
- Scaling a deployment or statefulset and restarting a rollout, which can be undone and are allowed by default, and deleting a pod, which waits for your approval by default. exec, attach, port-forward, applying manifests and editing other fields are never offered. Whatever the ServiceAccount’s Role does not allow is refused by the cluster itself.
- Which AI assistants can use Kubernetes through Flowplane?
- Claude, Claude Code, ChatGPT, Cursor, Codex, GitHub Copilot and Mistral Le Chat, and any other client that supports remote MCP servers with sign-in. They all reach Kubernetes through the same Flowplane address.
- What does it cost?
- Free for one person, with up to three apps and five assistants. Personal Pro is €9 a month including VAT; companies start at €99 a month for ten active users. Prices are indicative; see the pricing page.
Flowplane is not affiliated with or endorsed by Kubernetes’s maker; the name and logo only identify the app. Pricing
Let your AI work with your data. Safely, from today.
Free for yourself, set up in a minute. For your team or company, we’re happy to help you get started.