Kubernetes MCP Server by Flux159: Setup for Claude, Cursor and Other Apps

• Lists or reads Kubernetes resources

  Gets one resource or a list of resources of any kind, like kubectl get. Secret values are masked in the output.

  Try: “List the pods in the prod namespace.”

  Read · kubectl_get

• Describes a resource in detail

  Shows the detailed state and recent events of a resource, like kubectl describe.

  Try: “Describe the checkout deployment.”

  Read · kubectl_describe

• Reads logs from a pod or workload

  Returns container logs, like kubectl logs.

  Try: “Show the last logs from the api pod.”

  Read · kubectl_logs

• Explains what a resource field means

  Looks up the documentation for a resource type or field in your cluster's API.

  Try: “Explain the fields of a Deployment spec.”

  Read · explain_resource

• Lists the kinds your cluster supports

  Returns the API resource types available in the cluster.

  Try: “Which resource kinds does this cluster have?”

  Read · list_api_resources

• Switches or shows kubectl contexts

  Lists contexts or changes the active one, so later calls go to a different cluster.

  Try: “Switch to the staging cluster.”

  Write · kubectl_context

• Reconnects to the cluster API

  Rebuilds the API connection, for example after a cluster upgrade changed addresses.

  Try: “Reconnect to the cluster.”

  Read · kubectl_reconnect

• Checks the cluster connection

  Verifies that the server can reach your cluster.

  Try: “Is my cluster reachable?”

  Read · ping

• Applies a YAML manifest

  Creates or updates resources from a manifest, like kubectl apply.

  Try: “Apply this deployment YAML.”

  Write · kubectl_apply

• Creates resources

  Creates resources, like kubectl create.

  Try: “Create a namespace called demo.”

  Write · kubectl_create

• Edits fields of a resource

  Updates specific fields of an existing resource with a patch.

  Try: “Set the image of the api deployment to v2.”

  Write · kubectl_patch

• Changes the replica count

  Scales a deployment or similar resource up or down.

  Try: “Scale checkout to 5 replicas.”

  Write · kubectl_scale

• Manages deployment rollouts

  Checks status, restarts, pauses, resumes or undoes a rollout.

  Try: “Restart the api deployment.”

  Write · kubectl_rollout

• Deletes Kubernetes resources

  Deletes any resource, like kubectl delete.

  Try: “Delete the old demo namespace.”

  Delete · kubectl_delete

• Runs any kubectl command

  Runs a kubectl command you or the assistant write, so it can do anything kubectl can.

  Try: “Run kubectl top nodes.”

  Delete · kubectl_generic

• Runs a command inside a pod

  Runs a command in a container and returns the output. It runs the program directly, without a shell, and is marked destructive.

  Try: “Run ls -la /app in the api pod.”

  Runs code · exec_in_pod

• Installs a Helm chart

  Installs a chart into the cluster with optional values, repository and version.

  Try: “Install the nginx chart in the web namespace.”

  Write · install_helm_chart

• Upgrades a Helm release

  Upgrades an installed release to a new chart version or values.

  Try: “Upgrade the nginx release to the latest chart.”

  Write · upgrade_helm_chart

• Removes a Helm release

  Uninstalls a release from the cluster.

  Try: “Uninstall the nginx release.”

  Delete · uninstall_helm_chart

• Forwards a pod or service port to your computer

  Starts a port forward from a pod or service to a local port.

  Try: “Forward port 5432 from the postgres service.”

  Write · port_forward

• Stops a port forward

  Ends a port forward the server started.

  Try: “Stop forwarding the postgres port.”

  Write · stop_port_forward

• Cordons, uncordons or drains a node

  Marks a node unschedulable or schedulable, or drains it by evicting its pods.

  Try: “Drain node-3 for maintenance.”

  Delete · node_management

• Cleans up resources the server manages

  Removes the resources the server itself manages. The server treats it as destructive and turns it off in non-destructive mode.

  Try: “Clean up what the server created.”

  Delete · cleanup

Watch out for

• It can delete resources and run arbitrary kubectl

  Delete, node drain, Helm uninstall, pod exec and a generic kubectl tool are all on by default. Turn on read-only mode, or at least non-destructive mode. github.com

• It can switch your kubectl context

  A context tool can point later commands at another cluster, such as production. Check the active context before you approve changes. github.com

• Optional network mode has no sign-in

  The SSE and streamable HTTP modes are named unsafe by the project. Use the default local mode. github.com

• Secret masking is limited

  Secrets are masked in kubectl get output, but the README says this does not cover logs. github.com

Security advisories