RELATED 01
MCP Server Kubernetes
Manage a Kubernetes cluster through kubectl and Helm from an MCP client
1.6kGitHub stars
26k/wknpm downloads
80popularity score
Kubernetes MCP Server
Manage Kubernetes and OpenShift clusters from an assistant through the cluster's API
By Containers Go Local · stdio DevOps and monitoringCloud platforms
88 popularity score ★ 2.2k ↓ 69k/wk
Connect your client
Local (npm)
Add in one click
Opens the app, which asks before it adds the server.
Or add it by command or config
Claude CodeTerminal
claude mcp add kubernetes -- npx -y kubernetes-mcp-server@latestClaude Desktopclaude_desktop_config.json
{
"mcpServers": {
"kubernetes": {
"command": "npx",
"args": [
"-y",
"kubernetes-mcp-server@latest"
]
}
}
}Cursor~/.cursor/mcp.json
{
"mcpServers": {
"kubernetes": {
"command": "npx",
"args": [
"-y",
"kubernetes-mcp-server@latest"
]
}
}
}VS Code.vscode/mcp.json
{
"servers": {
"kubernetes": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"kubernetes-mcp-server@latest"
]
}
}
}Codex~/.codex/config.toml
[mcp_servers.kubernetes]
command = "npx"
args = ["-y", "kubernetes-mcp-server@latest"]Gemini CLITerminal
gemini mcp add --scope user kubernetes npx -- -y kubernetes-mcp-server@latestLocal (PyPI)
Add in one click
Opens the app, which asks before it adds the server.
Or add it by command or config
Claude CodeTerminal
claude mcp add kubernetes -- uvx kubernetes-mcp-server@latestClaude Desktopclaude_desktop_config.json
{
"mcpServers": {
"kubernetes": {
"command": "uvx",
"args": [
"kubernetes-mcp-server@latest"
]
}
}
}Cursor~/.cursor/mcp.json
{
"mcpServers": {
"kubernetes": {
"command": "uvx",
"args": [
"kubernetes-mcp-server@latest"
]
}
}
}VS Code.vscode/mcp.json
{
"servers": {
"kubernetes": {
"type": "stdio",
"command": "uvx",
"args": [
"kubernetes-mcp-server@latest"
]
}
}
}Codex~/.codex/config.toml
[mcp_servers.kubernetes]
command = "uvx"
args = ["kubernetes-mcp-server@latest"]Gemini CLITerminal
gemini mcp add --scope user kubernetes uvx -- kubernetes-mcp-server@latestSettings 7
KUBECONFIG |
Optional: Path to the kubeconfig, read by the Kubernetes client library when no kubeconfig is set in TOML. |
|---|---|
MCP_CONFIG_PATH |
Optional: Path to the main TOML file (read_only, toolsets, port and so on); ignored when --config is given. |
KUBE_CLIENT_QPS |
Optional: Request rate limit for the Kubernetes API client; 0 uses the client's own default. |
KUBE_CLIENT_BURST |
Optional: Burst limit for the Kubernetes API client; 0 uses the client's own default. |
| Set together: telemetry | |
OTEL_EXPORTER_OTLP_ENDPOINT |
Optional: OpenTelemetry collector endpoint; setting it turns on trace and metric export. |
OTEL_EXPORTER_OTLP_PROTOCOL |
Optional: OTLP protocol, grpc or http/protobuf. |
OTEL_TRACES_SAMPLER |
Optional: Trace sampling strategy, such as traceidratio. |
What it does
Kubernetes MCP Server connects an MCP client to one or more Kubernetes or OpenShift clusters. It is a Go program that calls the Kubernetes API server directly instead of wrapping kubectl or helm, so neither has to be installed. It finds clusters through your kubeconfig, which it watches for changes, or through in-cluster credentials when it runs in a pod.
Only the core and config toolsets load by default. Together they cover everyday cluster work: listing and reading pods and their logs, running commands in containers, starting pods, events, namespaces and OpenShift projects, node logs and usage figures, and get, list, apply, delete and scale on any resource by apiVersion and kind. Optional toolsets add Helm releases, kcp workspaces, Kiali and Istio mesh views and config edits, KubeVirt virtual machines, NetObserv network flows and Tekton pipelines. Recent releases add four observability toolsets: Loki log queries (observability/logs), Prometheus and Alertmanager queries (observability/metrics), OpenTelemetry Collector config schemas and validation (observability/otelcol) and Tempo trace search (observability/traces). The tool snapshots at the release tag show 77 tools with every toolset on. There are also 16 prompts, such as a cluster health check.
It speaks stdio unless a port is set in TOML, then serves Streamable HTTP at /mcp, where it can require OAuth/OIDC. Settings live in TOML: read_only, disable_destructive, toolsets, enabled_tools, disabled_tools and denied_resources trim what it exposes, and confirmation_rules can ask the user before risky calls in clients that support elicitation. Its config code caps one tool result at 16 MiB (max_result_bytes), one backend response at 4 MiB (max_backend_response_bytes) and runs 64 calls at once (max_in_flight). It ships as native binaries for Linux, macOS and Windows, an npm package, a PyPI package, a container image and a Helm chart.
What it needs
| Runtime | Node.jsPythonA prebuilt binary |
|---|---|
| Runs on | macOS, Windows, Linux |
| Kubernetes or OpenShift cluster | Your own instanceA kubeconfig (~/.kube/config or $KUBECONFIG) or an in-cluster service account; it acts with that identity's RBAC rights |
| Also |
|
Pick one way to run it: Node.js for npx, uv for uvx, or the native release binary, which needs neither. vm_guest_info needs the QEMU guest agent running inside the VM. For production its Kubernetes setup guide suggests a dedicated ServiceAccount with read-only rights.
Limits
- Not covered Only core and config load by default; every other toolset must be named in the TOML toolsets setting
- Limit Runtime settings are TOML or a few environment variables; the only flags are --config, --config-dir and --version
- Limit An unknown key in a TOML file stops the server from starting
- Limit --config - reads the main TOML from stdin, but only in HTTP mode
- Size cap A tool result is capped at 16 MiB and a backend response at 4 MiB by default (max_result_bytes, max_backend_response_bytes)
- Rate limit At most 64 tool, prompt and resource calls run at once by default (max_in_flight)
- Not covered OAuth sign-in only applies in HTTP mode, which needs port set in TOML
- Limit The published container image serves Streamable HTTP on port 8080, not stdio
- Read only read_only keeps only read tools and disable_destructive drops destructive ones; both are off by default
- Limit api_resources_list is listed in the README on main but is not in the released package yet
Only the core and config toolsets load by default; Helm, kcp, Kiali, KubeVirt, NetObserv and Tekton tools must be enabled in the TOML toolsets setting. Runtime settings are TOML or a few environment variables, not command-line flags, and unknown TOML keys stop the server from starting. OAuth sign-in only works in HTTP mode. The published container image serves HTTP on port 8080 rather than stdio.
Who it suits
Suits developers and platform engineers who want an assistant to inspect and operate clusters they already reach with kubectl credentials, including OpenShift and multi-cluster kubeconfigs, without installing kubectl or helm. Because it acts with whatever rights the kubeconfig grants and can delete, exec and apply, it fits best with a dedicated ServiceAccount and read_only or disable_destructive turned on, as its own Kubernetes setup guide suggests for production. It is a community project under the containers organisation, not a release of the Kubernetes project or a cloud vendor; teams that only need a managed cloud's own Kubernetes console may prefer that vendor's server.
Tools 60 of 77
44 only read, 11 change things, 5 can delete or overwrite. A ? marks an optional parameter.
configuration_contexts_list- Lists the contexts in the kubeconfig and the API server each one points at
targets_list- With non-kubeconfig cluster providers such as ACM, lists the targets; renamed after the provider, e.g. cluster_list
configuration_view- Prints the kubeconfig in use as YAML, optionally trimmed to the current context
minified?: boolean events_list- Lists cluster events in all namespaces or one, with paging
context?: stringcontinue?: stringfieldSelector?: stringlimit?: integernamespace?: string namespaces_list- Lists namespaces, with paging and field selectors
context?: stringcontinue?: stringfieldSelector?: stringlimit?: integer projects_list- Lists OpenShift projects
context?: stringcontinue?: stringlimit?: integer nodes_log- Reads kubelet, kube-proxy or another log file from a node through the API proxy
context?: stringname: stringquery: stringtailLines?: integer nodes_stats_summary- Returns a node's CPU, memory, disk and network statistics from the kubelet Summary API
context?: stringname: string nodes_top- Shows CPU and memory use per node from Metrics Server
context?: stringcontinue?: stringlabel_selector?: stringlimit?: integername?: string pods_list- Lists pods across all namespaces, filtered by label or field
context?: stringcontinue?: stringfieldSelector?: stringlabelSelector?: stringlimit?: integer pods_list_in_namespace- Lists pods in one namespace
context?: stringcontinue?: stringfieldSelector?: stringlabelSelector?: stringlimit?: integernamespace: string pods_get- Fetches a single pod by name
context?: stringname: stringnamespace?: string pods_deletedeletes- Deletes a pod by name
context?: stringname: stringnamespace?: string pods_top- Shows CPU and memory use per pod from Metrics Server
all_namespaces?: booleancontext?: stringcontinue?: stringlabel_selector?: stringlimit?: integername?: stringnamespace?: string pods_execdeletes- Runs a command in a pod's container and returns the output
command: arraycontainer?: stringcontext?: stringname: stringnamespace?: string pods_log- Reads a container's logs, including the previous instance
container?: stringcontext?: stringname: stringnamespace?: stringprevious?: booleantail?: integer pods_runchanges- Starts a pod from a container image, optionally exposing a port
context?: stringimage: stringname?: stringnamespace?: stringport?: number resources_list- Lists objects of any apiVersion and kind, by namespace, label or field
apiVersion: stringcontext?: stringcontinue?: stringfieldSelector?: stringkind: stringlabelSelector?: stringlimit?: integernamespace?: string resources_get- Fetches one object of any kind by name
apiVersion: stringcontext?: stringkind: stringname: stringnamespace?: string resources_create_or_updatedeletes- Applies a YAML or JSON manifest, creating or replacing the object
context?: stringresource: string resources_deletedeletes- Deletes an object of any kind, with an optional grace period
apiVersion: stringcontext?: stringgracePeriodSeconds?: integerkind: stringname: stringnamespace?: string resources_scalechanges- Reads or sets the replica count of a Deployment, StatefulSet or other scalable object
apiVersion: stringcontext?: stringkind: stringname: stringnamespace?: stringscale?: integer helm_installchanges- Installs a Helm chart as a release, with optional values
chart: stringname?: stringnamespace?: stringvalues?: object helm_list- Lists Helm releases in one namespace or all of them
all_namespaces?: booleanlimit?: integernamespace?: stringoffset?: integer helm_uninstalldeletes- Removes a Helm release
name: stringnamespace?: string vm_clonechanges- Copies a KubeVirt virtual machine under a new name
name: stringnamespace: stringtargetName: string vm_console_screenshot- Takes a picture of a virtual machine's VNC console
name: stringnamespace: string vm_createchanges- Creates a KubeVirt virtual machine from a workload, size or instance type
autostart?: booleaninstancetype?: stringname: stringnamespace: stringnetworks?: arrayperformance?: stringpreference?: stringsize?: stringstorage?: stringworkload?: string vm_guest_info- Reads operating system, network and disk details from inside a VM via the guest agent
info_type?: enumname: stringnamespace: string vm_lifecyclechanges- Starts, stops or restarts a virtual machine
action: enumname: stringnamespace: string vm_create_from_templatechanges- Creates a virtual machine from a VirtualMachineTemplate with parameters
namespace: stringparameters?: objecttemplate_name: string vm_troubleshoot- Collects a VM's status, conditions and events to help diagnose it
name: stringnamespace: string loki_list_instances- Finds the Loki instances in the cluster
loki_label_names- Lists the log label names a Loki instance holds
end?: stringlokiName?: stringlokiNamespace?: stringstart?: stringtenant?: string loki_label_values- Lists the values seen for one Loki label
end?: stringlabel: stringlokiName?: stringlokiNamespace?: stringstart?: stringtenant?: string loki_query_range- Runs a LogQL query over a time range and returns log lines
direction?: stringduration?: stringend?: stringlimit?: integerlokiName?: stringlokiNamespace?: stringquery: stringstart?: stringtenant?: string list_metrics- Lists Prometheus metric names that match a regular expression
name_regex: string execute_instant_query- Runs a PromQL query at one point in time
query: stringtime?: string execute_range_query- Runs a PromQL query over a time range at a set step
duration?: stringend?: stringquery: stringstart?: stringstep: string show_timeseries- Runs a PromQL range query and returns it for display as a chart
description?: stringduration?: stringend?: stringquery: stringstart?: stringstep: stringtitle?: string get_label_names- Lists Prometheus label names, optionally for one metric
end?: stringmetric?: stringstart?: string get_label_values- Lists the values of one Prometheus label
end?: stringlabel: stringmetric?: stringstart?: string get_series- Finds the time series that match given selectors
end?: stringmatches: stringstart?: string get_alerts- Lists Alertmanager alerts, filtered by state or receiver
active?: booleanfilter?: stringinhibited?: booleanreceiver?: stringsilenced?: booleanunprocessed?: boolean get_silences- Lists Alertmanager silences
filter?: string otelcol_list_components- Lists OpenTelemetry Collector receivers, processors, exporters and other components for a version
version?: string otelcol_get_component_schema- Returns the configuration schema of one Collector component
component_name: stringcomponent_type: stringversion?: string otelcol_validate_config- Checks a Collector component's configuration against its schema
component_name: stringcomponent_type: stringconfig: stringformat?: stringversion?: string tempo_list_instances- Finds the Tempo instances in the cluster
tempo_get_trace_by_id- Fetches one trace from Tempo by its ID
end?: stringstart?: stringtempoName: stringtempoNamespace: stringtenant?: stringtraceid: string tempo_search_traces- Searches Tempo for traces with a TraceQL query
end?: stringlimit?: integerquery: stringspss?: integerstart?: stringtempoName: stringtempoNamespace: stringtenant?: string tempo_search_tags- Lists the span and resource attribute names Tempo has seen
end?: stringlimit?: integermaxStaleValues?: integerquery?: stringscope?: stringstart?: stringtempoName: stringtempoNamespace: stringtenant?: string tempo_search_tag_values- Lists the values seen for one trace attribute
end?: stringlimit?: integermaxStaleValues?: integerquery?: stringstart?: stringtag: stringtempoName: stringtempoNamespace: stringtenant?: string tekton_pipeline_startchanges- Starts a Tekton Pipeline by creating a PipelineRun with parameters
name: stringnamespace?: stringparams?: object tekton_pipelinerun_lifecyclechanges- Restarts or cancels a PipelineRun
action: enumname: stringnamespace?: string tekton_pipelinerun_logs- Reads the logs of the TaskRuns in a PipelineRun
name: stringnamespace?: stringstep?: stringtail?: integertask?: string tekton_pipelinerun_diagnose- Collects a failed PipelineRun's conditions, failed steps with log tails, and warning events in one call
name: stringnamespace: string tekton_task_startchanges- Starts a Tekton Task by creating a TaskRun
name: stringnamespace?: stringparams?: object tekton_taskrun_restartchanges- Runs a TaskRun again with the same spec
name: stringnamespace?: string tekton_taskrun_logs- Reads a TaskRun's step logs from its pod
name: stringnamespace?: stringstep?: stringtail?: integer
Prompts 16
cluster-health-check- Walks through a health check of a Kubernetes or OpenShift cluster, optionally one namespace
mesh-list-applications- Lists applications in the Istio mesh namespaces (kiali)
list-istio-config- Lists Istio configuration objects in the mesh namespaces (kiali)
mesh-list-namespaces- Lists namespaces with their sidecar injection state (kiali)
mesh-list-services- Lists services in the mesh namespaces (kiali)
mesh-list-workloads- Lists workloads in the mesh namespaces (kiali)
mesh-health-check- Checks control plane and data plane health of the mesh (kiali)
mesh-topology- Describes mesh control plane components and cluster links (kiali)
traffic-topology- Maps service dependencies and traffic flow in the mesh (kiali)
service-troubleshoot- Looks into service errors through logs, traces and Istio config (kiali)
trace-analysis- Looks for slow spans and error sources in a service's traces (kiali)
istio-config-review- Reviews Istio configuration in a namespace for mistakes (kiali)
vm-troubleshoot- Produces step-by-step troubleshooting for a KubeVirt VM (kubevirt)
windows-golden-image- Guides building a Windows golden image with the windows-efi-installer Tekton pipeline (kubevirt)
hco-status- Reports the state of the HyperConverged Cluster Operator (kubevirt)
pipeline-troubleshoot- Gathers a PipelineRun's status, failed step logs and events for troubleshooting (tekton)