AUTOMATIONSWITCH

kubectl MCP Server

by Rohit Ghumare (CNCF Ambassador)

CNCF Landscape-listed community kubectl MCP server. 253 tools spanning kubectl, Helm, RBAC audit, cost optimisation, network diagnostics, and a 3D cluster topology UI. Multi-distribution via npm, PyPI, Docker Hub, and GHCR.

877·7 tools·Released MAR 2025·MIT
npx -y kubectl-mcp-server
Share:

The widest-surface community kubectl MCP server in the directory. 253 tools, 8 prompts, 8 resources, dashboard mode, RBAC audit, cost analysis, and a 3D cluster topology UI shipped in v1.24.0. CNCF Landscape listed, maintained by a CNCF Ambassador. Pick this when you want one MCP server that covers diagnostics, cost optimisation, RBAC audit, and visual dashboards on top of any kube-context. Pick the containers/kubernetes-mcp-server alongside it for OpenShift workloads, or the Flux159 server for kubectl-and-Helm-only TypeScript shops. The 253-tool surface is both the strength and the trade-off: agents working in smaller context windows will struggle to tool-pick efficiently, so deploy with --read-only and an explicit subset for production workflows.

Reviewed by M. Nouriel · MAY 2026

INSTALL THIS SERVER

Requires authenticationInherits caller kubeconfig context. For production, use --read-only and a dedicated ServiceAccount with least-privilege RoleBindings.
{ "mcpServers": { "kubectl": { "command": "npx", "args": [ "-y", "kubectl-mcp-server" ] } } }
PrereqRequires Node.js 18+ and a valid kubeconfig (default ~/.kube/config). Add `--read-only` to args for production. Distributed on npm `kubectl-mcp-server`, PyPI `kubectl-mcp-server`, Docker Hub `rohitghumare64/kubectl-mcp-server`, GHCR. Path: ~/Library/Application Support/Claude/claude_desktop_config.json (macOS).
{ "mcpServers": { "kubectl": { "command": "npx", "args": [ "-y", "kubectl-mcp-server" ] } } }
{ "mcpServers": { "kubectl": { "command": "npx", "args": [ "-y", "kubectl-mcp-server" ] } } }
{ "mcpServers": { "kubectl": { "command": "npx", "args": [ "-y", "kubectl-mcp-server" ] } } }
{ "mcpServers": { "kubectl": { "command": "npx", "args": [ "-y", "kubectl-mcp-server" ] } } }

7 TOOLS AVAILABLE

get_pods
List pods across namespaces
Read
describe_pod
Surface events, logs, resource limits for a pod
Read
analyze_pod_crashes
AI-assisted crash diagnostics for failing pods
Read
audit_rbac
RBAC permission review across namespaces
Read
helm_install
Deploy a Helm chart
Admin
optimize_resources
Cost optimisation recommendations for cluster resources
Read

OUR ASSESSMENT

Strengths
  • Vast tool surface (253 tools) covering kubectl, Helm, RBAC audit, cost analysis, network diagnostics, dashboards.
  • CNCF Landscape inclusion provides an independent quality signal beyond GitHub stars.
  • Multi-distribution: npm `kubectl-mcp-server`, PyPI, Docker Hub `rohitghumare64/kubectl-mcp-server`, GHCR.
  • 3D cluster topology UI (v1.24.0) for human exploration of cluster state.
  • OAuth 2.1, RBAC validation, non-destructive mode, secret masking documented as enterprise features.
  • 24 releases shipped over 13 months.
  • 234 passing tests reported in the repo.
Weaknesses
  • 253 tools is a context-window risk on smaller models; agents will struggle to tool-pick efficiently with the full surface enabled.
  • 1 commit on the default branch in the last 30 days; release cadence has slowed.
  • Two open issues at time of review flag a missing destructive-command guard (#82) and a `timezone` import bug (#81), both with slow triage.
  • Single-maintainer governance offset partially by maintainer CNCF Ambassador credibility; key-person risk remains.
  • Some claimed integrations (browser automation, dashboard UI) are gated behind extras and pull a heavier dependency tree.
Security Notes

Inherits the caller kubeconfig context. Run with --read-only and a dedicated ServiceAccount with least-privilege RoleBindings for production. Prefer in-cluster deployment so kubeconfig credentials stay inside the cluster boundary. The 253-tool surface widens the blast radius if write tools are enabled; subset the toolset to the workflow at hand.

Best For

Platform teams running agentic Kubernetes operations who need maximum tool breadth in one server, operators comfortable scoping access via --read-only flag, and teams using AI clients beyond Claude (Cursor, Windsurf, Copilot, 15+ supported clients).

TECHNICAL DETAILS

Language
python
Transport
stdio
Clients
Claude DesktopClaude CodeCursorVS CodeWindsurf
License
MIT
npm
kubectl-mcp-server
Last Release
v1.24.0FEB 20, 2026
First Released
MAR 23, 2025

ADOPTION METRICS

// GitHub Stars
877

// Reading this877 stars and 167 forks. CNCF Landscape inclusion provides an independent editorial signal beyond stars.

// Popularity Rank
#15
Globally · #5 in Infrastructure

// Reading thisFifth-ranked in infrastructure category. The 253-tool kitchen-sink approach contrasts with the API-direct (containers/) and TypeScript-kubectl (Flux159) options already in the directory.

SOURCES & VERIFICATION

We don't take any single directory's word for it. Before scoring, we cross-reference 5 public MCP sources, install the server ourselves against the clients we cover, and record when we last re-verified.

01
Discovered
Manual submission
First indexed MAY 1, 2026
02
Cross-referenced
5 directories
PulseMCP, MCP.so, Glama, Smithery, Official MCP Registry
03
Verified against
Claude Desktop, Cursor, Windsurf, Copilot, VS Code
Installed and tested across clients
04
Last re-checked
MAY 1, 2026
Weekly re-verification
// How other directories see it

The same server, 5 different lenses. We reconcile these signals into our editorial score, which is why our number sometimes diverges from a directory-aggregate star count.

SourceTheir ratingTheir star countTheir downloadsLast synced
AutomationSwitch This page4.2editorial877MAY 1, 2026
PulseMCP— unratedunavailableunavailableMAY 1, 2026
MCP.so— unratedunavailableunavailableMAY 1, 2026
Glama— unratedunavailableunavailableMAY 1, 2026
Smithery— unratedunavailableunavailableMAY 1, 2026
Official MCP Registry— unratedunavailableunavailableMAY 1, 2026

// Counts are directory-reported; we don't adjust them. Discrepancies usually come from different snapshot times or star-caching.

OTHER INFRASTRUCTURE MCP SERVERS

Community4.5

Kubernetes MCP Server

containers/ org (Red Hat-adjacent)

Native Go MCP server for Kubernetes and OpenShift from the containers/ org. Talks directly to the Kubernetes API server, supports Helm and Tekton, ships with OpenTelemetry observability built in. Distributed as a single native binary plus npm and PyPI packages.

6 tools1,502
Vendor4.5

Sentry MCP Server

Sentry

Sentry's official MCP service. Runs as remote infrastructure at mcp.sentry.dev and ships a Claude Code plugin that auto-delegates Sentry questions. Designed for human-in-the-loop coding agents working through Sentry errors, traces, and performance data.

6 tools671
Vendor4.5

Azure AKS MCP

Microsoft Azure

Official Microsoft Azure MCP server for Azure Kubernetes Service. Workload Identity with federated tokens, three access-level guardrails (readonly, readwrite, admin), and a unified two-tool surface (call_az + call_kubectl). Top commit cadence in this batch (10 commits on main in the last 30 days).

6 tools132
Community4.3

mcp-server-kubernetes (Flux159)

Flux159 (community)

Community-favourite TypeScript Kubernetes MCP. Uses local kubectl plus Helm v3, distributed via npm, and ships as an mcpb extension for one-click Claude Desktop install. Mature 3.x release stream with consistent monthly cadence.

5 tools1,382
Vendor4.3

Argo CD MCP

Argo Project Labs (donated by Akuity)

Argoproj-Labs-blessed Argo CD MCP server, originally donated by Akuity. Both stdio and HTTP-stream transports out of the box, MCP_READ_ONLY mode for production safety, and a tool surface covering the full Argo CD application lifecycle (list/get/sync/delete plus resource-tree traversal and workload logs).

8 tools416
Vendor3.8

GKE MCP Server

GoogleCloudPlatform org

Google Kubernetes Engine MCP server in the GoogleCloudPlatform org. Adds GKE-specific tools (cluster management, upgrade risk reports, log queries) plus Gemini CLI extension support. Maintainers explicitly state this is for education and experimentation and is NOT FOR PRODUCTION USE.

8 tools147
// Get in touch

DISCUSS YOUR
MCP REQUIREMENTS.

Evaluating a server, scoping an internal deployment, or working out whether MCP is the right fit at all. Start the conversation and we will point you at the right piece of the ecosystem.

Discuss Your MCP Requirements →