MCP Kubectl
Examples are illustrative; names, counts, tool responses and agent wording vary. Check the actual result in your intended cluster.
Not ready
Find out why something is not ready.
A deployment looks stuck and you do not know whether the pods, the copies or the machines are the cause.
Illustrative conversation
YouIs the web deployment healthy, and are the nodes ready?
Tool callkubectl_deploy_status { "deployment": "web", "namespace": "shop" }
What happensreplicas 1, ready_replicas 1, updated_replicas 1, available_replicas 1, status healthy.
Tool callkubectl_cluster_health {}
What happenstotal_nodes 1, ready_nodes 1, not_ready_nodes 0, status healthy.
AgentThe web deployment has its one replica ready and available, and the one node is ready. These counts alone do not establish application health.
Those counts describe one snapshot; they do not prove every workload, controller or dependency is healthy.
What runs
See what is running in a namespace.
You want to know how much lives in a part of the cluster, and which pods are up.
Illustrative conversation
YouHow many pods, services and deployments are in shop, and which pods are not running?
Tool callkubectl_resource_summary { "namespace": "shop" }
What happensdeployments 4, pods 9, services 5.
Tool callkubectl_get_pod_status { "namespace": "shop" }
What happenspod_count 9 and one entry per pod, for example api-7c9d8f5b6c-m4h2p, ready 1/1, restarts 0, status Running, and one finished job with status Succeeded and ready 0/1.
AgentThere are 9 pods in 4 deployments with 5 services. The running ones are ready; the 0/1 entry is a finished job that Succeeded.
One pod
Look at one pod.
One pod is the suspect and you want its state without the rest.
Illustrative conversation
YouHow is the web pod doing?
Tool callkubectl_get_pod_status { "namespace": "shop", "pod_name": "web-5d8f7c9b6d-x2k4q" }
What happensready 2/2, restarts 0, status Running.
Tool callkubectl_get_pod_status { "namespace": "shop", "pod_name": "nonexistent" }
What happensTool error: pods "nonexistent" not found, reason NotFound, code 404.
AgentThe web pod is Running with both containers ready and no restarts. The second name does not exist, so that was my typo.
Namespace trap
Ask without a namespace and read the answer carefully.
The agent leaves out which part of the cluster you mean, and the answer looks too small or says nothing was found.
Illustrative conversation
YouCount what is running.
Tool callkubectl_resource_summary {}
What happensdeployments 0, pods 0, services 1, with no namespace field in the result: only default was counted.
Tool callkubectl_deploy_status { "deployment": "web" }
What happensTool error: deployments.apps "web" not found.
AgentThe summary used default; deployment status used the configured namespace. I will request shop explicitly for both.
Change and verify
Make a change and verify it.
You reviewed a file that creates a deployment and want to apply it.
Illustrative conversation
YouI applied docs-example.yaml in staging. Is the deployment up?
What happensYou ran: MCP Kubectl cli apply -f ./docs-example.yaml -n staging --dry-run server, then the same command without the dry run.
Tool callkubectl_deploy_status { "deployment": "docs-example", "namespace": "staging" }
What happensreplicas 2, ready_replicas 2, updated_replicas 2, available_replicas 2, status healthy.
Agentdocs-example has both replicas ready. To compare it with your file, run: MCP Kubectl cli get deployment docs-example -n staging -o yaml.
See change and verify.
Node maintenance
Take a node out of service.
A machine in the cluster needs maintenance inside an agreed window.
Illustrative conversation
YouI am draining worker-1 for maintenance. Watch the cluster until I am done.
What happensYou ran: MCP Kubectl cli cordon worker-1, then MCP Kubectl cli drain worker-1.
Tool callkubectl_cluster_health {}
What happenstotal_nodes 3, ready_nodes 3, not_ready_nodes 0: a cordoned node is still Ready, only unschedulable.
AgentThe cluster is healthy. Run MCP Kubectl cli uncordon worker-1 when the work is done and I will check again.
See maintain a node.