Kubernetes MCP Server 🚀

Created By

briankscheong8 months ago

The Kubernetes MCP Server is a Model Context Protocol (MCP) server that provides seamless integration with Kubernetes APIs, enabling advanced automation and interaction capabilities for developers, operators, and AI tools.

# go

# kubernetes

Overview Content Tools Comments

Content

Kubernetes MCP Server 🚀

Kubernetes MCP Server 🚀

Overview 📊

This MCP server enables AI tools to interact with Kubernetes clusters using natural language, providing capabilities to:

🔍 Retrieve and analyze cluster resources
📈 Monitor deployments, pods, and services
🛠️ Execute common kubectl operations through AI interfaces
🔧 Troubleshoot cluster issues with AI assistance

Prerequisites ✅

A Kubernetes cluster with API access
Valid kubeconfig file or service account credentials
Appropriate RBAC permissions for desired operations

Installation 💻

Usage with Claude Desktop

Add the following to your Claude Desktop configuration file (~/Library/Application Support/Claude/claude_desktop_config.json on macOS or %APPDATA%\Claude\claude_desktop_config.json on Windows):

{
  "mcpServers": {
    "kubernetes": {
      "command": "path/to/k8smcp",
      "args": [
        "stdio",
        "--kubeconfig=/path/to/your/kubeconfig"
      ]
    }
  }
}

Usage with VS Code

Add the following to your VS Code User Settings (JSON) file or .vscode/mcp.json in your workspace:

{
  "mcp": {
    "inputs": [
      {
        "type": "promptString",
        "id": "kubeconfig_path",
        "description": "Path to kubeconfig file",
        "default": "${env:HOME}/.kube/config"
      }
    ],
    "servers": {
      "kubernetes": {
        "command": "path/to/k8smcp",
        "args": [
          "stdio",
          "--kubeconfig=/path/to/your/kubeconfig"
        ]
      }
    }
  }
}

Usage with Cline

Add the following to your Cline configuration file (path/to/cline_mcp_settings.json after selecting "Configure MCP Servers"):

{
  "mcpServers": {
    "kubernetes": {
      "disabled": false,
      "timeout": 60,
      "command": "path/to/k8smcp",
      "args": [
        "stdio",
        "--read-only=false"
        "--kubeconfig=/path/to/your/kubeconfig"
      ],
      "env": {
        "K8S_MCP_TOOLSETS": "all"
      },
      "transportType": "stdio"
    }
  }
}

Make sure to update the command value with the path to your k8smcp executable. You can set the server configurations either using args or env.

Build from source

Clone the repository and build the binary:

git clone https://github.com/briankscheong/k8s-mcp-server.git
cd k8s-mcp-server
make build

Or install directly with Go:

go install github.com/briankscheong/k8s-mcp-server/cmd/k8s-mcp-server@latest

Command Line Options ⌨️

A Kubernetes MCP Server that provides tools for interacting with Kubernetes clusters.

Environment Variables:
  K8S_MCP_KUBECONFIG            Path to kubeconfig file
  K8S_MCP_NAMESPACE             Default Kubernetes namespace
  K8S_MCP_IN_CLUSTER            Use in-cluster config (true/false)
  K8S_MCP_READ_ONLY             Restrict to read-only operations (true/false)
  K8S_MCP_RESOURCE_TYPES        Comma-separated list of resource types
  K8S_MCP_TOOLSETS              Comma-separated list of toolsets to enable
  K8S_MCP_EXPORT_TRANSLATIONS   Export translations (true/false)

Usage:
  k8smcp [command]

Available Commands:
  completion  Generate the autocompletion script for the specified shell
  help        Help about any command
  sse         Start sse server
  stdio       Start stdio server

Flags:
      --export-translations      Save translations to a JSON file
  -h, --help                     help for k8smcp
      --in-cluster               Use in-cluster config instead of kubeconfig file
      --kubeconfig string        Path to the kubeconfig file (default "/Users/briancheong/.kube/config")
      --namespace string         Default Kubernetes namespace to target (default "default")
      --read-only                Restrict operations to read-only (no create, update, delete) (default true)
      --resource-types strings   Comma separated list of Kubernetes resource types to enable (pods,deployments,services,configmaps,namespaces,nodes) (default [all])
      --toolsets strings         Comma separated list of tools to enable (default [all])
  -v, --version                  version for k8smcp

Use "k8smcp [command] --help" for more information about a command.

Server Transport Options 🔄

stdio

The stdio transport is the default and recommended option for most users for local integration:

k8smcp stdio --kubeconfig=/path/to/your/kubeconfig

SSE

The sse transport provides support for HTTP-based JSON-RPC message transport. This can be helpful when deploying the server in a Kubernetes cluster that needs to expose a port for client connection.

k8smcp sse --in-cluster=true

NOTE

The --in-cluster=true flag needs to be set if the server is deployed in a Kubernetes cluster.

Access Control 🔒

By default, the server applies the permissions of the provided kubeconfig or service account. For enhanced security, you can:

Create a dedicated service account with restricted RBAC permissions
Set namespace limits to prevent cross-namespace operations
Enable read-only mode to prevent mutations to cluster state

Tools 🧰

The Kubernetes MCP Server provides a comprehensive set of tools for interacting with your Kubernetes cluster.

Resource Operations 📦

get_pod - Get detailed information about a specific pod
- namespace: Pod namespace (string, optional, defaults to current namespace)
- name: Pod name (string, required)
list_pods - List pods in a namespace
- namespace: Namespace to list pods from (string, optional, defaults to current namespace)
- label_selector: Filter pods by label selector (string, optional)
- field_selector: Filter pods by field selector (string, optional)
get_pod_logs - Get logs from a pod
- namespace: Pod namespace (string, optional, defaults to current namespace)
- name: Pod name (string, required)
- container: Container name (string, optional, defaults to first container)
- tail_lines: Number of lines to retrieve from the end (number, optional)
- previous: Get logs from previous container instance (boolean, optional)
get_deployment - Get information about a specific deployment
- namespace: Deployment namespace (string, optional, defaults to current namespace)
- name: Deployment name (string, required)
list_deployments - List deployments in a namespace
- namespace: Namespace to list deployments from (string, optional, defaults to current namespace)
- label_selector: Filter deployments by label selector (string, optional)
get_service - Get information about a specific service
- namespace: Service namespace (string, optional, defaults to current namespace)
- name: Service name (string, required)
list_services - List services in a namespace
- namespace: Namespace to list services from (string, optional, defaults to current namespace)
- label_selector: Filter services by label selector (string, optional)
get_configmap - Get information about a specific ConfigMap
- namespace: ConfigMap namespace (string, optional, defaults to current namespace)
- name: ConfigMap name (string, required)
list_configmaps - List ConfigMaps in a namespace
- namespace: Namespace to list ConfigMaps from (string, optional, defaults to current namespace)
- label_selector: Filter ConfigMaps by label selector (string, optional)
list_namespaces - List all namespaces in the cluster
- No parameters required
list_nodes - List all nodes in the cluster
- No parameters required

Management Operations ⚙️

delete_pod - Delete a pod from a namespace
- namespace: Pod namespace (string, optional, defaults to current namespace)
- name: Pod name (string, required)
- grace_period_seconds: Grace period before deletion (number, optional)
scale_deployment - Scale a deployment to a specific number of replicas
- namespace: Deployment namespace (string, optional, defaults to current namespace)
- name: Deployment name (string, required)
- replicas: Number of replicas (number, required)

IMPORTANT

By default, tools that involve modification of resources in the cluster are disabled. To enable them, you have to set the --read-only=false flag or the K8S_MCP_READ_ONLY=false environment variable.