K8s Mcp Server

Kubernetes MCP Server

A Kubernetes Model Context Protocol (MCP) server that provides tools for interacting with Kubernetes clusters through a standardized interface.

Hosted deployment

A hosted deployment is available on Fronteir AI.

Features

• API Resource Discovery: Get all available API resources in your Kubernetes cluster.
• Resource Listing: List resources of any type with optional namespace and label filtering.
• Resource Details: Get detailed information about specific Kubernetes resources.
• Resource Description: Get comprehensive descriptions of Kubernetes resources, similar to kubectl describe.
• Pod Logs: Retrieve logs from specific pods (optionally from a specific container, or all containers if unspecified).
• Node Metrics: Get resource usage metrics for specific nodes.
• Pod Metrics: Get CPU and Memory metrics for specific pods.
• Event Listing: List events within a namespace or for a specific resource.
• Resource Creation/Updating: Create new Kubernetes resources or update existing ones from a YAML or JSON manifest.
• Resource Deletion: It deletes a resource in the Kubernetes cluster based on the provided namespace and kind.
• Standardized Interface: Uses the MCP protocol for consistent tool interaction.
• Flexible Configuration: Supports different Kubernetes contexts and resource scopes.
• Multiple Modes: Run in stdio mode for CLI tools, sse mode, or streamable-http mode for web applications, and --readonly for no change in the cluster.
• Security: Runs as non-root user in Docker containers for enhanced security.

Prerequisites

• Go 1.23 or later
• Access to a Kubernetes cluster
• kubectl configured with appropriate cluster access

Installation

1. Clone the repository:

``bash git clone https://github.com/reza-gholizade/k8s-mcp-server.git cd k8s-mcp-server ``

1. Install dependencies:

``bash go mod download ``

1. Build the server:

``bash go build -o k8s-mcp-server main.go ``

Usage

Starting the Server

The server can run in three modes, configurable via command-line flags or environment variables.

Stdio Mode (for CLI integrations)

This mode uses standard input/output for communication.

./k8s-mcp-server --mode stdio

Or using environment variables:

SERVER_MODE=stdio ./k8s-mcp-server

SSE Mode (for web applications)

This mode starts an HTTP server with Server-Sent Events support.

Default (port 8080):

./k8s-mcp-server --mode sse

Specify a port:

./k8s-mcp-server --mode sse --port 9090

Or using environment variables:

SERVER_MODE=sse SERVER_PORT=9090 ./k8s-mcp-server

Streamable-HTTP Mode (for web applications)

This mode starts an HTTP server with streamable-http transport support, following the MCP specification.

Default (port 8080):

./k8s-mcp-server --mode streamable-http

Specify a port:

./k8s-mcp-server --mode streamable-http --port 9090

Or using environment variables:

SERVER_MODE=streamable-http SERVER_PORT=9090 ./k8s-mcp-server

The server will be available at http://localhost:8080/mcp (or your specified port).

If no mode is specified, it defaults to SSE on port 8080.

Kubernetes Authentication

The server supports multiple authentication methods, which are tried in the following order of priority:

1. Kubeconfig Content from Environment Variable

You can provide the entire kubeconfig file content via the KUBECONFIG_DATA environment variable:

export KUBECONFIG_DATA="$(cat ~/.kube/config)"
./k8s-mcp-server

This is useful when you want to avoid mounting files or when running in environments where file access is restricted.

2. API Server URL and Token

You can authenticate using a Kubernetes API server URL and bearer token:

export KUBERNETES_SERVER="https://kubernetes.example.com:6443"
export KUBERNETES_TOKEN="your-bearer-token-here"
./k8s-mcp-server

Optional environment variables for TLS configuration:

• KUBERNETES_CA_CERT: CA certificate content (base64-encoded or PEM format)
• KUBERNETES_CA_CERT_PATH: Path to CA certificate file
• KUBERNETES_INSECURE: Set to "true" to skip TLS verification (not recommended for production)

Example with CA certificate:

export KUBERNETES_SERVER="https://kubernetes.example.com:6443"
export KUBERNETES_TOKEN="your-bearer-token-here"
export KUBERNETES_CA_CERT_PATH="/path/to/ca.crt"
./k8s-mcp-server

3. In-Cluster Authentication (Service Account)

When running inside a Kubernetes cluster, the server automatically detects and uses the service account token from /var/run/secrets/kubernetes.io/serviceaccount/token. This is the recommended method for running the server as a pod within a cluster.

Example Deployment:

apiVersion: v1
kind: ServiceAccount
metadata:
  name: k8s-mcp-server-sa
  namespace: default
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  name: k8s-mcp-server-role
rules:
  - apiGroups: [""]
    resources: ["*"]
    verbs: ["get", "list", "watch", "create", "update", "patch", "delete"]
  - apiGroups: ["apps"]
    resources: ["*"]
    verbs: ["get", "list", "watch", "create", "update", "patch", "delete"]
  # Add more rules as needed for your use case
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: k8s-mcp-server-rb
subjects:
  - kind: ServiceAccount
    name: k8s-mcp-server-sa
    namespace: default
roleRef:
  kind: ClusterRole
  name: k8s-mcp-server-role
  apiGroup: rbac.authorization.k8s.io
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: k8s-mcp-server
  namespace: default
spec:
  replicas: 1
  selector:
    matchLabels:
      app: k8s-mcp-server
  template:
    metadata:
      labels:
        app: k8s-mcp-server
    spec:
      serviceAccountName: k8s-mcp-server-sa
      containers:
        - name: k8s-mcp-server
          image: ginnux/k8s-mcp-server:latest
          ports:
            - containerPort: 8080
          env:
            - name: SERVER_MODE
              value: "sse"
            - name: SERVER_PORT
              value: "8080"

4. Kubeconfig File Path (Default)

If none of the above methods are available, the server falls back to using a kubeconfig file:

• Uses the path provided via --kubeconfig flag (if implemented) or KUBECONFIG environment variable
• Defaults to ~/.kube/config if neither is specified

# Using default ~/.kube/config
./k8s-mcp-server

# Using custom kubeconfig path
export KUBECONFIG=/path/to/your/kubeconfig
./k8s-mcp-server

Note: The server automatically detects which authentication method to use based on the available environment variables and file system. You don't need to explicitly configure the authentication method - it will use the first available method in the priority order listed above.

Read-Only Mode

The server supports a read-only mode that disables all write operations, providing a safer way to explore and monitor your Kubernetes cluster without the risk of making changes.

Enable read-only mode with the --read-only flag:

./k8s-mcp-server --read-only

You can combine read-only mode with any server mode:

# Read-only with stdio mode
./k8s-mcp-server --mode stdio --read-only

# Read-only with SSE mode
./k8s-mcp-server --mode sse --read-only

# Read-only with streamable-http mode
./k8s-mcp-server --mode streamable-http --read-only

When read-only mode is enabled, the following tools are disabled:

• createResource (Kubernetes resource creation/updates)
• helmInstall (Helm chart installations)
• helmUpgrade (Helm chart upgrades)
• helmUninstall (Helm chart uninstallations)
• helmRollback (Helm release rollbacks)
• helmRepoAdd (Helm repository additions)

All other read-only operations remain available, including listing resources, getting logs, viewing metrics, and inspecting Helm releases.

Tool Category Flags

You can selectively disable entire categories of tools using these flags:

Disable Kubernetes Tools:

./k8s-mcp-server --no-k8s

Disable Helm Tools:

./k8s-mcp-server --no-helm

Combine with other flags:

# Read-only mode with only Kubernetes tools (no Helm)
./k8s-mcp-server --read-only --no-helm

# Read-only mode with only Helm tools (no Kubernetes)
./k8s-mcp-server --read-only --no-k8s

# SSE mode with only Kubernetes tools
./k8s-mcp-server --mode sse --no-helm

Note: You cannot use both --no-k8s and --no-helm together, as this would result in no available tools. The server will exit with an error if both flags are provided.

When --no-k8s is enabled, all Kubernetes tools are disabled:

• getAPIResources, listResources, getResource, describeResource
• getPodsLogs, getNodeMetrics, getPodMetrics, getEvents
• createResource (if not in read-only mode)

When --no-helm is enabled, all Helm tools are disabled:

• helmList, helmGet, helmHistory, helmRepoList
• helmInstall, helmUpgrade, helmUninstall, helmRollback, helmRepoAdd (if not in read-only mode)

Using the Docker Image

You can also run the server using the pre-built Docker image from Docker Hub.

1. Pull the image:

``bash docker pull ginnux/k8s-mcp-server:latest ` You can replace latest with a specific version tag (e.g., 1.0.0`).

1. Run the container:

Note: The server supports multiple authentication methods. You can either mount a kubeconfig file (as shown below) or use environment variables for authentication (see [Kubernetes Authentication](#kubernetes-authentication) section above).

• SSE Mode (default behavior of the image):

``bash docker run -p 8080:8080 -v ~/.kube/config:/home/appuser/.kube/config:ro ginnux/k8s-mcp-server:latest ` This maps port 8080 of the container to port 8080 on your host and mounts your Kubernetes config read-only to the non-root user's home directory. The server will be available at http://localhost:8080. The image defaults to sse mode on port 8080`.

• Streamable-HTTP Mode:

``bash docker run -p 8080:8080 -v ~/.kube/config:/home/appuser/.kube/config:ro ginnux/k8s-mcp-server:latest --mode streamable-http ` This runs the server in streamable-http mode. The server will be available at http://localhost:8080/mcp`.

• Stdio Mode:

``bash docker run -i --rm -v ~/.kube/config:/home/appuser/.kube/config:ro ginnux/k8s-mcp-server:latest --mode stdio ` The -i flag is important for interactive stdio communication. --rm` cleans up the container after exit.

• Custom Port for SSE Mode:

``bash docker run -p 9090:9090 -v ~/.kube/config:/home/appuser/.kube/config:ro ginnux/k8s-mcp-server:latest --mode sse --port 9090 ``

• Custom Port for Streamable-HTTP Mode:

``bash docker run -p 9090:9090 -v ~/.kube/config:/home/appuser/.kube/config:ro ginnux/k8s-mcp-server:latest --mode streamable-http --port 9090 ``

• Alternative: Mount entire .kube directory:

``bash docker run -p 8080:8080 -v ~/.kube:/home/appuser/.kube:ro ginnux/k8s-mcp-server:latest ``

• Using environment variables for authentication (no file mounting required):

```bash # Using kubeconfig content from environment variable docker run -p 8080:8080 \ -e KUBECONFIG_DATA="$(cat ~/.kube/config)" \ ginnux/k8s-mcp-server:latest

# Or using API server URL and token docker run -p 8080:8080 \ -e KUBERNETESSERVER="https://kubernetes.example.com:6443" \ -e KUBERNETESTOKEN="your-token-here" \ -e KUBERNETESCACERT_PATH="/path/to/ca.crt" \ -v /path/to/ca.crt:/path/to/ca.crt:ro \ ginnux/k8s-mcp-server:latest ```

Using with Docker Compose

Create a docker-compose.yml file:

Option 1: Using kubeconfig file (traditional method):

version: '3.8'
services:
  k8s-mcp-server:
    image: ginnux/k8s-mcp-server:latest # Or a specific version
    container_name: k8s-mcp-server
    ports:
      - "8080:8080" # Host:Container, adjust if using a different SERVER_PORT
    volumes:
      - ~/.kube:/home/appuser/.kube:ro # Mount kubeconfig read-only to non-root user home
    environment:
      - KUBECONFIG=/home/appuser/.kube/config
      - SERVER_MODE=sse # Can be 'stdio', 'sse', or 'streamable-http'
      - SERVER_PORT=8080 # Port for SSE/streamable-http modes
    # command: ["--read-only"] # Uncomment this line to enable read-only mode
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8080/"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 10s

Option 2: Using environment variables (no file mounting):

version: '3.8'
services:
  k8s-mcp-server:
    image: ginnux/k8s-mcp-server:latest
    container_name: k8s-mcp-server
    ports:
      - "8080:8080"
    environment:
      - KUBECONFIG_DATA=${KUBECONFIG_DATA} # Set this in your .env file or shell
      # Or use API server and token:
      # - KUBERNETES_SERVER=https://kubernetes.example.com:6443
      # - KUBERNETES_TOKEN=${KUBERNETES_TOKEN}
      - SERVER_MODE=sse
      - SERVER_PORT=8080
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8080/"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 10s

Note: To enable read-only mode, use the command override as shown in Option 1. For stdio mode, you might need to adjust 'ports', add 'stdin_open: true' and 'tty: true', and potentially override the command.

Then start with:

docker compose up -d

To see logs: docker compose logs -f k8s-mcp-server.

Security Considerations

The Docker image runs as a non-root user (appuser with UID 1001) for enhanced security:

• The application binary is located at /usr/local/bin/k8s-mcp-server
• The kubeconfig should be mounted to /home/appuser/.kube/config
• Health checks are enabled to monitor container status
• The container includes minimal dependencies (ca-certificates and curl only)

Making API Calls (SSE/Streamable-HTTP Mode)

Once the server is running in SSE or streamable-http mode, you can make JSON-RPC calls to its HTTP endpoint:

curl -X POST -H "Content-Type: application/json" -d '{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "getAPIResources",
    "arguments": {
      "includeNamespaceScoped": true,
      "includeClusterScoped": true
    }
  }
}' http://localhost:8080/

You can also check the health status:

curl -f http://localhost:8080/

Available Tools

1. getAPIResources

Retrieves all available API resources in the Kubernetes cluster.

Parameters:

• includeNamespaceScoped (boolean, optional): Whether to include namespace-scoped resources (defaults to true).
• includeClusterScoped (boolean, optional): Whether to include cluster-scoped resources (defaults to true).

Example:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "getAPIResources",
    "arguments": {
      "includeNamespaceScoped": true,
      "includeClusterScoped": true
    }
  }
}

2. listResources

Lists all instances of a specific resource type.

Parameters:

• Kind (string, required): The kind of resource to list (e.g., "Pod", "Deployment").
• namespace (string, optional): The namespace to list resources from. If omitted, lists across all namespaces for namespaced resources (subject to RBAC).
• labelSelector (string, optional): Filter resources by label

…

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: reza-gholizade
• Source: reza-gholizade/k8s-mcp-server
• License: MIT

Install and usage instructions live in the source repository linked above.