cleanup
Cleanup all managed resources
LABSCO SUMMARY
An agent can operate the cluster, not only read it.
What it takes off you is the terminal round trip: the verbs you would have typed — get, describe, logs, apply, scale, rollout, drain — arrive as tools, so the investigation and the fix it implies happen in the same conversation. How far that goes is yours to set: read-only, non-destructive, or an explicit tool list, decided at startup rather than per call.
What it is
A kubectl and Helm surface over a cluster. Most calls mirror a kubectl verb one for one, and a generic call falls through to any kubectl command for what the named ones do not cover. Values from kubectl get secrets come back masked unless you turn that off.
What you get
- Get and describe any resource type, in one namespace or across all of them, filtered by label or field selector
- Apply, create, patch, delete and scale — from a manifest string or a file, with a dry run available on several of them
- Logs from pods, deployments and jobs: tailed, followed, since a timestamp, or from the previous container
- Rollout management on deployments, daemonsets and statefulsets, including a rollback to a named revision
- Cordon, drain and uncordon on nodes, with a dry run and a grace period
- Helm install, upgrade and uninstall, from a repo chart or a template, with values inline or from a file
- A port forward to a resource, and a separate call to stop it again
- A command run inside a pod, passed as an argument array rather than a shell string
- The API resources the cluster exposes, and the reference documentation for a resource or a single field
- Context listing, reading and switching, and a reconnect that rebuilds every API client after a control-plane upgrade rotates addresses
- A fall-through to any kubectl command with your own arguments and flags
- A k8s-diagnose prompt that walks a pod problem from a keyword and an optional namespace through to a remediation suggestion
- Three ways to narrow what the agent can do: ALLOW_ONLY_READONLY_TOOLS, ALLOW_ONLY_NON_DESTRUCTIVE_TOOLS for read plus create and update, or ALLOWED_TOOLS as an explicit list
Requirements
npx on your PATH, kubectl installed, and a cluster kubectl can already reach — kubectl get pods succeeding in a terminal is the check. It follows your current context by default; KUBECONFIG_PATH points at a different file, and KUBECONFIG_YAML or KUBECONFIG_JSON pass the config inline for containers with no file to mount. Secrets are masked in responses unless MASK_SECRETS is set to false.
Setup effort
One command — npx mcp-server-kubernetes
USE CASES
01
Working out why a pod will not start
Describe the pod, read its logs including the previous container, and check what the events say — in one thread.
02
Shipping a change and watching it land
Apply the manifest, follow the rollout, and roll back to a named revision if it goes wrong.
03
Taking a node out for maintenance
Cordon it, drain it with a grace period, do the work, then uncordon — with a dry run first.
04
Reaching a service that has no ingress
Forward a local port to the pod or service, use it, then stop the forward by id rather than hunting the process.
05
Installing or upgrading a chart
Install with your own values, upgrade in place later, and uninstall the release when it is finished with.
06
Letting an agent look but not touch
Start it with ALLOW_ONLY_READONLY_TOOLS — the write verbs never appear in the tool list it is offered.
23 tools
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
exec_in_pod
Execute a command in a Kubernetes pod or container and return the output. Command must be an array of strings where the first element is the executable and remaining elements are arguments. This executes directly without shell interpretation for security.
explain_resource
Get documentation for a Kubernetes resource or field
install_helm_chart
Install a Helm chart with support for both standard and template-based installation
kubectl_apply
Apply a Kubernetes YAML manifest from a string or file
kubectl_context
Manage Kubernetes contexts - list, get, or set the current context
kubectl_create
Create Kubernetes resources using various methods (from file or using subcommands)
kubectl_delete
Delete Kubernetes resources by resource type, name, labels, or from a manifest file
kubectl_describe
Describe Kubernetes resources by resource type, name, and optionally namespace
kubectl_generic
Execute any kubectl command with the provided arguments and flags
kubectl_get
Get or list Kubernetes resources by resource type, name, and optionally namespace
kubectl_logs
Get logs from Kubernetes resources like pods, deployments, or jobs
kubectl_patch
Update field(s) of a resource using strategic merge patch, JSON merge patch, or JSON patch
kubectl_reconnect
Reconnect to the Kubernetes API server by recreating all API clients. Use this after cluster upgrades (e.g., EKS control plane upgrades that rotate ENIs/IPs) to force fresh DNS resolution and new TCP connections.
kubectl_rollout
Manage the rollout of a resource (e.g., deployment, daemonset, statefulset)
kubectl_scale
Scale a Kubernetes deployment
list_api_resources
List the API resources available in the cluster
node_management
Manage Kubernetes nodes with cordon, drain, and uncordon operations
ping
Verify that the counterpart is still responsive and the connection is alive.
port_forward
Forward a local port to a port on a Kubernetes resource
stop_port_forward
Stop a port-forward process
uninstall_helm_chart
Uninstall a Helm chart release
upgrade_helm_chart
Upgrade an existing Helm chart release
BEFORE YOU INSTALL
What to know first
Tools you get
23 tools when we connected. cleanup_pods, helm_template_apply and helm_template_uninstall appear in the documentation but not on the running server; clean up pods with kubectl_get and kubectl_delete using field selectors.
Requires
- npx on PATH
- kubectl on PATH
- A kubeconfig with a context for a cluster kubectl can already reach
CONTEXT COST
Moderate Standing cost
Rated by the size of the 23 tool definitions that ride along on every turn, not by how many there are
INSTALLATION
Connect your client
Server setup
TERMINAL
npx mcp-server-kubernetesMCP Clients
Claude Code
CLAUDE CODE · TERMINAL
claude mcp add kubernetes -- npx mcp-server-kubernetesCodex
CODEX · TERMINAL
codex mcp add kubernetes -- npx mcp-server-kubernetesClaude Desktop, Cursor
JSON
{
"mcpServers": {
"kubernetes": {
"command": "npx",
"args": [
"mcp-server-kubernetes"
]
}
}
}Claude Desktop — extension
Settings -> Extensions -> Browse Extensions and scroll to find mcp-server-kubernetes
VS Code
VS CODE · JSON
{
"servers": {
"kubernetes": {
"command": "npx",
"args": [
"mcp-server-kubernetes"
],
"description": "Kubernetes cluster management and operations"
}
}
}MCP Chat
MCP CHAT · TERMINAL
npx mcp-chat --server "npx mcp-server-kubernetes"
