CommunityActive· 8d★ 4.3by Flux159

Kubernetes

Inspect Kubernetes clusters from Claude. Pods, deployments, logs, port-forward suggestions. Sane defaults for SREs.

View repo

Safety & Trust

Delv Safety Grade: C

Score 58/100 · assessed 2026-04-28

Maintainer45

Permissions35

Supply chain65

Transparency70

Incidents100

This community MCP server provides read-heavy Kubernetes cluster inspection through Claude, wrapping kubectl-style operations. The maintainer Flux159 appears to be a solo developer with limited GitHub presence, creating meaningful bus factor risk for production use. Permissions are the primary concern: full kubeconfig access grants read and potentially write capabilities across entire clusters, including secrets, configmaps, and all namespace resources. The server reads your local kubeconfig file, inheriting whatever RBAC permissions that context holds. While the stated intent is read-only inspection (logs, pod lists, deployments), the underlying Kubernetes API client has no enforced restrictions. Supply chain is reasonable via npm packaging. Transparency is adequate with open source code and basic documentation. No known security incidents. Suitable for local development clusters but requires careful RBAC scoping for production access.

Lethal Trifecta (prompt-injection exposure)

TWO OF THREE

Private dataYes

Reads secrets, credentials, private files

Untrusted inputNo

Ingests web pages, PRs, issues, emails

External commsYes

Can send data outbound

Cluster control with kubeconfig is private; outbound to the API server. Pod logs can carry untrusted strings if you read them — push to 3/3 for log-reading workloads.

Green flags

Open source with clear repository structure and reasonable documentation
Published via npm with standard package distribution
Focused use case for SRE diagnostics rather than cluster mutation
No known security incidents or malicious behaviour
Transparent about Kubernetes API interaction model

Red flags

Solo maintainer with limited public track record and low commit activity
Full kubeconfig access inherits all RBAC permissions without restriction
Can read cluster secrets and sensitive configmaps if kubeconfig permits
No enforced read-only mode despite stated inspection-focused use case
Filesystem read access to kubeconfig may expose multiple cluster credentials

Permissions requested

Read filesOutbound networkPrivate networkDB readDB writeAccess secrets

Assessed by Delv Editorial using public metadata. Grades are advisory and update as the ecosystem changes. They do not replace your own review of permissions and code before granting an agent access to sensitive systems.

Review

This MCP server turns Claude into a kubectl-adjacent interface for your Kubernetes clusters. Point it at a kubeconfig and you can ask Claude to list pods, describe deployments, tail logs, or suggest port-forward commands without leaving the chat. It's built for SREs who want conversational cluster inspection rather than another dashboard. I'd reach for this when I'm debugging a flaky service and want to correlate logs across pods without writing kubectl one-liners. Ask "show me pods in the production namespace that restarted in the last hour" and it translates that into the right API calls. The port-forward suggestions are particularly handy: Claude will tell you the exact command to run locally, which saves the mental overhead of remembering syntax. The tool respects your existing kubeconfig contexts, so it works with whatever clusters you already have configured. No new auth dance. It exposes read-only operations by default, which is the right call for an LLM-driven tool. You're not going to accidentally delete a deployment by asking a vague question. Quirks: it's community-maintained, so expect rough edges. The repo doesn't spell out every supported Kubernetes resource type, and I'd test it against your specific cluster version before relying on it in anger. Error messages when a namespace doesn't exist or a pod name is ambiguous could be clearer. Also, if you're running a massive cluster with hundreds of pods, responses can get verbose. Claude will try to summarise, but you'll want to narrow your queries. Skip this if you're not already comfortable with Kubernetes concepts. It won't teach you what a StatefulSet is, and it assumes you know what you're looking for. Also skip if you need write operations: this is strictly for inspection and diagnostics. If your workflow is "check Grafana, then kubectl exec into the pod", this slots in nicely. If you're doing GitOps deployments or Helm wrangling, it won't help.

Verdict

Install this if you're an SRE or platform engineer who spends time in kubectl and wants a conversational layer for cluster diagnostics. Skip it if you need write access to your clusters or you're still learning Kubernetes basics. It's a time-saver for people who already know what they're doing.

Good at

Respects your existing kubeconfig and contexts, so no new authentication setup.
Port-forward suggestions save you looking up syntax every time.
Read-only by default means you won't accidentally break things by asking the wrong question.
Useful for correlating logs across multiple pods without writing shell scripts.
Works well for quick diagnostics when you're already in a Claude chat.

Watch out

Community-maintained, so updates and support depend on the author's availability.
Error handling for ambiguous queries or missing resources could be more helpful.
Large clusters produce verbose responses that can overwhelm the chat interface.
No write operations, so you'll still need kubectl for deployments or config changes.
Documentation doesn't enumerate every supported Kubernetes resource type.

Getting started

1. Install via npm: `npm install -g mcp-server-kubernetes`. Requires Node.js 18 or later. 2. Add the server to your Claude Desktop config (usually `~/Library/Application Support/Claude/claude_desktop_config.json` on macOS). Include the path to your kubeconfig if it's not in the default location. 3. Restart Claude Desktop. Ask "list all pods in the default namespace" to verify it's connected. 4. Watch out for verbose output if you have large clusters. Narrow your queries by namespace or label selector to keep responses manageable. 5. Check the repo's README for the full list of supported commands, as not every kubectl operation is exposed.

Works with

Claude DesktopClaude CodeCursor