Kubernetes MCP Server

Community

Native Kubernetes and OpenShift control for AI assistants via MCP.

How to install ↓ Compare GitHub source repository ↗

Category: Dev Tools #74 of 438
Stars: ★ 2.2k Very popular
Transport: stdio (local process) · Streamable HTTP
Runtime: Node.js · Python · Docker · Prebuilt binary
Credentials: No credential needed
License: Apache-2.0
Last commit: 4 days ago
Tools: 46

63FMRS · C

The README describes an actively maintained, well-documented Go-native MCP server with broad tool coverage (Kubernetes, OpenShift, Helm, Tekton, KubeVirt, Kiali, NetObserv) and multiple install paths (npm, PyPI/uvx, native binaries, Docker); no third-party user reviews are included in the source material.

Strongest · Documentation 17/20 Weakest · Reliability 10/20

Reliability

10/20

Security and permissions

10/20

Maintenance

14/20

Documentation

17/20

Setup experience

12/20

Why each score

Reliability 10/20: The README claims an 'extensive test suite' and the repo carries a build.yaml CI status badge, indicating some CI pipeline exists; the documented tool list is internally consistent with the described feature set with no obvious contradictions. However, this is a static-only review — no test files, coverage, or workflow contents were inspected, and there is no way to verify that any of the three packaging methods (npm, PyPI, OCI) actually complete the MCP init + tool-list handshake in practice. All three packages in server.json carry a placeholder version of '0.0.0', which further undermines confidence that declared behavior matches real, published artifacts. Per the static calibration rule, without verifiable execution evidence the score cannot exceed 12, so this is scored 10.
Security and permissions 10/20: By default the server runs in read-write mode and exposes many high-risk operations: pods_exec (arbitrary command execution inside containers), pods_delete, resources_delete, resources_create_or_update (a full server-side-apply overwrite), helm_install/uninstall, resources_scale, and KubeVirt's vm_lifecycle — none of which have a server-enforced confirmation step by default; confirmation is left entirely to the client UI. Mitigations do exist but are opt-in: --read-only, --disable-destructive, a denied_resources TOML blocklist, and OAuth/OIDC support for HTTP mode; the docs also state that tokens, keys, passwords and cloud credentials are automatically redacted before being sent to clients, and a dedicated least-privilege ServiceAccount setup guide is referenced. No red-line issue was found (no malware, no covert exfiltration, no fabricated identity/results), but the broad default permission scope combined with destructive operations enabled by default and no built-in confirmation matches 'main risks visible but permissions/confirmation/scoping incomplete,' scored 10.
Maintenance 14/20: The repository belongs to the containers GitHub organization (home to Podman, Buildah, and related projects), is Apache-2.0 licensed, has 1957 stars, is not archived, and shows 93 open issues indicating ongoing community engagement. The docs directory is extensive and appears actively maintained (multiple topic-specific guides, an ecosystem validation table with eval scenario counts). However, all three package entries in server.json use a placeholder version '0.0.0', which either reflects a manifest-generation gap or makes it impossible to confirm actual release cadence and version discipline; no explicit security-response channel or release policy is stated. Given generally positive activity signals offset by governance gaps, scored 14.
Documentation 17/20: Documentation coverage is broad: beyond the README there is a dedicated docs/ directory covering configuration reference (configuration.md), observability (OTEL.md), logging (logging.md), custom prompts (prompts.md), OIDC setup guides (Keycloak, Entra ID), client-specific getting-started guides (Claude Code, a secure Kubernetes deployment guide), plus toolset-specific docs (KIALI.md, kubevirt.md, NETOBSERV.md). Tool parameters are auto-generated into clear tables listing each argument. The ecosystem validation table further links to checkable evidence (evals/tasks). The main gap is the absence of a consolidated 'known limitations' or explicitly labeled troubleshooting section. Scored 17.
Setup experience 12/20: Concrete configuration examples are provided for multiple mainstream clients (Claude Desktop, VS Code/Insiders, Cursor, Goose CLI, Claude Code), and npx/uvx enable near one-line startup. The Docker/OCI path requires manually mounting the kubeconfig, setting an env var, and mapping a port — documented but more manual. The stated prerequisite of 'access to a Kubernetes cluster' is an inherent domain dependency, not a setup flaw. However, all three packages in server.json list a placeholder version of '0.0.0', meaning a user following the manifest literally cannot be sure which actual version would be fetched — a real risk of setup instructions diverging from a genuinely installable artifact. Per the static calibration rule, without verified execution evidence the setup score cannot exceed 15, so this is scored 12.

Static review · not runListed 2026-08-26

Read the FMRS scoring method →

Fit and risk

What it can accessRuns commands or codeUses the network

Best for

Developers and platform engineers who want AI assistants (Claude Desktop, Claude Code, VS Code, Cursor) to interact directly with Kubernetes/OpenShift clusters
Teams needing lightweight, dependency-free cluster access without kubectl/helm binaries installed locally
Multi-cluster environments defined across kubeconfig contexts

Not for

Users without access to a Kubernetes cluster
Strictly read-only or isolated environments where --read-only/--disable-destructive flags haven't been reviewed first
Users who just want a thin kubectl wrapper (this is a native reimplementation, not kubectl passthrough)

Required permissions

Requires a kubeconfig file (or in-cluster ServiceAccount) with API access to the target cluster(s)
Docker/OCI deployment requires a mounted kubeconfig volume and an exposed HTTP port
By default has full read/write access to cluster resources unless --read-only or --disable-destructive is set
The pods_exec tool can run arbitrary commands inside cluster containers

Risks and side effects

Default configuration allows create/update/delete on any Kubernetes/OpenShift resource, including Secrets, unless restricted via denied_resources or read-only mode
pods_exec and resources_create_or_update can be used to run arbitrary commands or alter cluster state
Streamable HTTP mode exposes a network endpoint (port 8080 in the Docker example) that should be secured with the supported OAuth/OIDC before exposing beyond localhost
Multi-cluster support means a single compromised MCP session could reach every cluster defined in the kubeconfig

Setup

Before you start

Runtime：Node.js · Python · Docker · Prebuilt binary

Other optional settings (1)

KUBECONFIG optional Environment variable setting the kubeconfig path inside the container (typically /kubeconfig), used with a docker -v mount of the host's ~/.kube/config.

For Claude Desktop, edit claude_desktop_config.json to add: {"mcpServers":{"kubernetes":{"command":"npx","args":["-y","kubernetes-mcp-server@latest"]}}} (requires Node.js/npm). Alternatively run the PyPI package via uvx (requires Python/uv), or download a native Go binary release. For Docker/OCI, mount the kubeconfig file as a volume, set the KUBECONFIG env var, map port 8080, and connect over streamable-http to http://localhost:8080/mcp. VS Code, VS Code Insiders, and Cursor have one-click install links; Goose CLI uses a similar npx entry in config.yaml. Requires access to a Kubernetes cluster.

claude_desktop_config.json

{"mcpServers":{"kubernetes":{"command":"npx","args":["-y","kubernetes-mcp-server@latest"]}}}

Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json

{
  "servers": {
    "kubernetes": {
      "command": "npx",
      "args": [
        "-y",
        "kubernetes-mcp-server@latest"
      ]
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal

claude mcp add kubernetes -- npx -y kubernetes-mcp-server@latest

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

After connecting, the client's tool list should show tools like namespaces_list and pods_list; then ask the client to 'list all namespaces' — a returned namespace list confirms the connection works.

Troubleshooting

Verify the --kubeconfig path or in-cluster config resolves correctly if the server can't reach the cluster
Use a higher --log-level value for verbose Kubernetes API debugging
If a tool is missing, check that --toolsets includes the needed toolset (e.g., helm, tekton, kubevirt)
For Docker/OCI deployment, confirm the kubeconfig volume mount path matches the KUBECONFIG env var inside the container
Check the /stats endpoint or OpenTelemetry tracing (docs/OTEL.md) for observability issues

Things to try

Once connected, you can ask your AI assistant things like:

List all pods in Running phase across the cluster
Show the last 100 lines of logs for pod my-pod in the default namespace
Scale the apps/v1 Deployment my-app to 5 replicas
Install a Helm release named grafana from the stable/grafana chart into the monitoring namespace

Tools 46

configuration_contexts_list read-only

List all available context names and associated server URLs from the kubeconfig file

targets_list read-only

List all available targets

configuration_view read-only

Get the current Kubernetes configuration content as a kubeconfig YAML

events_list read-only

List Kubernetes events (warnings, errors, state changes) for debugging and troubleshooting

namespaces_list read-only

List all Kubernetes namespaces in the current cluster

projects_list read-only

List all OpenShift projects in the current cluster

nodes_log read-only

Get logs from a Kubernetes node (kubelet, kube-proxy, or other system logs) via the API proxy

nodes_stats_summary read-only

Get detailed resource usage statistics from a node via the kubelet Summary API, including PSI metrics

Show 38 more tools

nodes_top read-only

List CPU and memory consumption for nodes as recorded by the Kubernetes Metrics Server

pods_list read-only

List all Kubernetes pods in the current cluster from all namespaces

pods_list_in_namespace read-only

List all Kubernetes pods in the specified namespace

pods_get read-only

Get a Kubernetes Pod in the current or provided namespace by name

pods_delete destructive

Delete a Kubernetes Pod in the current or provided namespace by name

pods_top read-only

List CPU and memory consumption for pods as recorded by the Metrics Server

pods_exec writes

Execute a command in a Kubernetes Pod (shell access, run commands in container)

pods_log read-only

Get the logs of a Kubernetes Pod by name

pods_run writes

Run a Kubernetes Pod with the provided container image and optionally expose a port

resources_list read-only

List Kubernetes resources and objects by apiVersion and kind, optionally namespace and label selector

resources_get read-only

Get a Kubernetes resource by apiVersion, kind, namespace, and name

resources_create_or_update writes

Create or update a Kubernetes resource via Server-Side Apply using a complete desired-state manifest

resources_delete destructive

Delete a Kubernetes resource by apiVersion, kind, namespace, and name

resources_scale writes

Get or update the scale (replica count) of a Kubernetes resource

helm_install writes

Install a Helm chart to create a release in the current or provided namespace

helm_list read-only

List Helm releases in the current, provided, or all namespaces

helm_uninstall destructive

Uninstall a Helm release in the current or provided namespace

kcp_workspaces_list read-only

List all available kcp workspaces in the current cluster

kcp_workspace_describe read-only

Get detailed information about a specific kcp workspace

kiali_get_mesh_traffic_graph read-only

Return service-to-service traffic topology, dependencies, and network metrics for the specified namespaces

kiali_get_mesh_status read-only

Retrieve the high-level health, topology, and environment details of the Istio service mesh

kiali_manage_istio_config_read read-only

Read Istio, Gateway API, and Inference API configuration (list or get)

kiali_manage_istio_config writes

Create, patch, or delete Istio, Gateway API, and Inference API configuration

kiali_list_mesh_clusters read-only

List Istio mesh clusters accessible to Kiali

kiali_get_resource_details read-only

Fetch a list of resources or detailed data for a specific resource (apps, ArgoCD apps)

kiali_list_traces read-only

List distributed traces for a service in a namespace

kiali_get_trace_details read-only

Fetch a single distributed trace by trace ID and return its call hierarchy

kiali_get_pod_performance read-only

Return a summary of Pod CPU/memory usage compared to Kubernetes requests/limits

kiali_get_logs read-only

Get the logs of a Kubernetes Pod or workload in a namespace

kiali_get_metrics read-only

Return a compact summary of Istio metrics (latency, traffic, throughput, payload sizes) for a resource

vm_clone writes

Clone a VirtualMachine on KubeVirt by creating a VirtualMachineClone resource

vm_create writes

Create a VirtualMachine on KubeVirt with automatic instance type, preference, and image resolution

vm_guest_info read-only

Get guest OS information from a VirtualMachine's QEMU guest agent

vm_lifecycle writes

Manage KubeVirt VirtualMachine lifecycle: start, stop, or restart

vm_troubleshoot read-only

Diagnose KubeVirt VirtualMachine issues with automated root-cause detection

netobserv_list_flows read-only

List NetObserv network flow records from Loki

netobserv_get_flow_metrics read-only

Return aggregated NetObserv flow metrics as topology or time-series data

netobserv_export_flows read-only

Export NetObserv flow records as CSV with the same filters as list_flows

Use cases

Diagnose and automatically fix failing Deployments

Manage pods, namespaces, and generic Kubernetes/OpenShift resources

Install, list, and uninstall Helm releases

Start and troubleshoot Tekton Pipelines and Tasks

Create, clone, and troubleshoot KubeVirt virtual machines

Inspect Istio service mesh topology and metrics via Kiali

Analyze network flows and packet drops via NetObserv

Supported clients

Claude Desktop

Claude Code

VS Code

Cursor

Goose CLI

Listed from the project's documentation, not tested by this site.

Overview

Kubernetes MCP Server is a Go-native Model Context Protocol server that talks directly to the Kubernetes/OpenShift API server rather than wrapping kubectl or helm CLI tools. It supports CRUD on pods, generic resources, namespaces, and events, plus optional toolsets for Helm, Tekton, KubeVirt, Kiali (Istio service mesh), and NetObserv (network observability), with multi-cluster, read-only, and fine-grained resource-restriction support.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →

NPM Sentinel MCP Server 79 · B

AI-powered NPM package analysis MCP server

★ 18 · Tools 19 Compare with this →

MartinLoop MCP Server 79 · B

An independent control layer that gives AI coding agents budgets, brakes, and receipts

★ 413 Compare with this →

GitHub MCP Server 76 · B

GitHub's official MCP server for managing repos, issues, PRs, and workflows via natural language

★ 33.7k · Tools 15 Compare with this →

Compare with other MCP servers

Source revision 970df6a7169c Data synced 2026-10-19 Read the FMRS scoring method