Kubernetes MCP Server

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

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.
• 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.
• 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:

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

2. Install dependencies:

   go mod download
   

3. Build the server:

   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.

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:

   docker pull ginnux/k8s-mcp-server:latest
   

   You can replace latest with a specific version tag (e.g., 1.0.0).

2. Run the container:

   • SSE Mode (default behavior of the image):

     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:

     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:

     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:

     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:

     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:

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

Using with Docker Compose

Create a docker-compose.yml file:

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
    # To run in stdio mode with docker-compose, you might need to adjust 'ports',
    # add 'stdin_open: true' and 'tty: true', and potentially override the command.
    # For example, to force stdio mode:
    # command: ["--mode", "stdio"]
    # stdin_open: true
    # tty: true
    # For streamable-http mode, simply change SERVER_MODE to 'streamable-http'

To enable read-only mode, use the command override as shown above.

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 selector (e.g., "app=nginx,env=prod").

Example:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "listResources",
    "arguments": {
      "Kind": "Pod",
      "namespace": "default",
      "labelSelector": "app=nginx"
    }
  }
}

3. getResource

Retrieves detailed information about a specific resource.

Parameters:

• kind (string, required): The kind of resource to get (e.g., "Pod", "Deployment").
• name (string, required): The name of the resource to get.
• namespace (string, optional): The namespace of the resource (required for namespaced resources).

Example:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "getResource",
    "arguments": {
      "kind": "Pod",
      "name": "nginx-pod",
      "namespace": "default"
    }
  }
}

4. describeResource

Describes a resource in the Kubernetes cluster, similar to kubectl describe.

Parameters:

• Kind (string, required): The kind of resource to describe (e.g., "Pod", "Deployment").
• name (string, required): The name of the resource to describe.
• namespace (string, optional): The namespace of the resource (required for namespaced resources).

Example:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "describeResource",
    "arguments": {
      "Kind": "Pod",
      "name": "nginx-pod",
      "namespace": "default"
    }
  }
}

5. getPodsLogs

Retrieves the logs of a specific pod.

Parameters:

• Name (string, required): The name of the pod.
• namespace (string, required): The namespace of the pod.
• containerName (string, optional): The specific container name within the pod. If omitted:
  • If the pod has one container, its logs are fetched.
  • If the pod has multiple containers, logs from all containers are fetched and concatenated.

Example:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "getPodsLogs",
    "arguments": {
      "Name": "my-app-pod-12345",
      "namespace": "production",
      "containerName": "main-container"
    }
  }
}

6. getNodeMetrics

Retrieves resource usage metrics for a specific node.

Parameters:

• Name (string, required): The name of the node.

Example:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "getNodeMetrics",
    "arguments": {
      "Name": "worker-node-1"
    }
  }
}

7. getPodMetrics

Retrieves CPU and Memory metrics for a specific pod.

Parameters:

• namespace (string, required): The namespace of the pod.
• podName (string, required): The name of the pod.

Example:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "getPodMetrics",
    "arguments": {
      "namespace": "default",
      "podName": "my-app-pod-67890"
    }
  }
}

8. getEvents

Retrieves events for a specific namespace or resource.

Parameters:

• namespace (string, optional): The namespace to get events from. If omitted, events from all namespaces are considered (subject to RBAC).
• resourceName (string, optional): The name of a specific resource (e.g., a Pod name) to filter events for.
• resourceKind (string, optional): The kind of the specific resource (e.g., "Pod") if resourceName is provided.

Example (Namespace Events):

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "getEvents",
    "arguments": {
      "namespace": "default"
    }
  }
}

Example (Resource Events):

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "getEvents",
    "arguments": {
      "namespace": "production",
      "resourceName": "my-app-pod-12345",
      "resourceKind": "Pod"
    }
  }
}

9. createOrUpdateResource

Creates a new resource or updates an existing one from a YAML or JSON manifest.

Parameters:

• manifest (string, required): The YAML or JSON manifest of the resource.
• namespace (string, optional): The namespace in which to create/update the resource. If the manifest contains a namespace, this parameter can be used to override it. If not provided and the manifest doesn't specify one, "default" might be assumed or it might be an error depending on the resource type.

Example:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "createOrUpdateResource",
    "arguments": {
      "namespace": "default",
      "manifest": "apiVersion: v1\nkind: Pod\nmetadata:\n  name: my-new-pod\nspec:\n  containers:\n  - name: nginx\n    image: nginx:latest"
    }
  }
}

Helm Operations

10. helmInstall

Install a Helm chart to the Kubernetes cluster.

Parameters:

• releaseName (string, required): Name of the Helm release
• chartName (string, required): Name or path of the Helm chart
• namespace (string, optional): Kubernetes namespace for the release (defaults to "default")
• repoURL (string, optional): Helm repository URL
• values (object, optional): Values to override in the chart

Example:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "helmInstall",
    "arguments": {
      "releaseName": "my-nginx",
      "chartName": "bitnami/nginx",
      "namespace": "web",
      "repoURL": "https://charts.bitnami.com/bitnami",
      "values": {
        "replicaCount": 3,
        "service": {
          "type": "LoadBalancer"
        }
      }
    }
  }
}

11. helmUpgrade

Upgrade an existing Helm release.

12. helmUninstall

Uninstall a Helm release from the Kubernetes cluster.

13. helmList

List all Helm releases in the cluster or a specific namespace.

14. helmGet

Get details of a specific Helm release.

15. helmHistory

Get the history of a Helm release.

16. helmRollback

Rollback a Helm release to a previous revision.

Adding New Tools

1. Define the Tool: In tools/tools.go, define a function that returns an mcp.Tool structure. This includes the tool's name, description, and input/output schemas.
2. Implement the Handler: In handlers/handlers.go, create a handler function. This function takes *k8s.Client as an argument and returns a function with the signature func(context.Context, mcp.ToolInput) (mcp.ToolOutput, error). This inner function will contain the logic for your tool.
3. Register the Tool: In main.go, add your new tool to the MCP server instance using s.AddTool(tools.YourToolDefinitionFunction(), handlers.YourToolHandlerFunction(client)).

Contributing

Contributions are welcome! Please see CONTRIBUTING.md for details on how to contribute to this project.

License

[email protected]

This project is licensed under the MIT License - see the LICENSE file for details.

VS Code Integration

Quick Setup

Automatic Installation (Recommended)

macOS/Linux:

curl -sSL https://raw.githubusercontent.com/reza-gholizade/k8s-mcp-server/main/scripts/install-vscode-config.sh | bash

Windows (PowerShell):

iex ((New-Object System.Net.WebClient).DownloadString('https://raw.githubusercontent.com/reza-gholizade/k8s-mcp-server/main/scripts/install-vscode-config.ps1'))

Manual Installation

1. Install the MCP extension in VS Code:

   code --install-extension modelcontextprotocol.mcp
   

2. Add to your VS Code settings.json:

   Open VS Code settings (Cmd/Ctrl + ,) → Open Settings JSON → Add:

   macOS/Linux:

   {
     "mcp.mcpServers": {
       "k8s-mcp-server": {
         "command": "k8s-mcp-server",
         "args": ["--mode", "stdio"],
         "env": {
           "KUBECONFIG": "${env:HOME}/.kube/config"
         }
       }
     }
   }
   

   Read-Only Mode (recommended for safety):

   {
     "mcp.mcpServers": {
       "k8s-mcp-server": {
         "command": "k8s-mcp-server",
         "args": ["--mode", "stdio", "--read-only"],
         "env": {
           "KUBECONFIG": "${env:HOME}/.kube/config"
         }
       }
     }
   }
   

   Kubernetes Tools Only:

   {
     "mcp.mcpServers": {
       "k8s-mcp-server": {
         "command": "k8s-mcp-server",
         "args": ["--mode", "stdio", "--no-helm"],
         "env": {
           "KUBECONFIG": "${env:HOME}/.kube/config"
         }
       }
     }
   }
   

   Helm Tools Only:

   {
     "mcp.mcpServers": {
       "k8s-mcp-server": {
         "command": "k8s-mcp-server",
         "args": ["--mode", "stdio", "--no-k8s"],
         "env": {
           "KUBECONFIG": "${env:HOME}/.kube/config"
         }
       }
     }
   }
   

   Read-Only with Kubernetes Tools Only:

   {
     "mcp.mcpServers": {
       "k8s-mcp-server": {
         "command": "k8s-mcp-server",
         "args": ["--mode", "stdio", "--read-only", "--no-helm"],
         "env": {
           "KUBECONFIG": "${env:HOME}/.kube/config"
         }
       }
     }
   }
   

   Windows:

   {
     "mcp.mcpServers": {
       "k8s-mcp-server": {
         "command": "k8s-mcp-server.exe",
         "args": ["--mode", "stdio"],
         "env": {
           "KUBECONFIG": "${env:USERPROFILE}/.kube/config"
         }
       }
     }
   }
   

3. Ensure the binary is in your PATH:

   Download the appropriate binary from the releases page and add it to your system PATH.

4. Restart VS Code

Usage in VS Code

Once configured, you can use the Kubernetes MCP server in VS Code with Claude or other MCP-compatible tools:

1. Open VS Code
2. Access Claude (or other MCP-enabled AI assistant)
3. Use natural language to interact with your Kubernetes cluster:
   • "List all pods in the default namespace"
   • "Show me the logs for pod nginx-123"
   • "Get the CPU usage for worker-node-1"
   • "Describe the deployment called my-app"

Configuration Options

You can customize the configuration by modifying the settings:

{
  "mcp.mcpServers": {
    "k8s-mcp-server": {
      "command": "k8s-mcp-server",
      "args": ["--mode", "stdio"],
      "env": {
        "KUBECONFIG": "/path/to/your/kubeconfig",
        "KUBERNETES_CONTEXT": "your-context-name"
      }
    }
  }
}

Troubleshooting

• Binary not found: Ensure k8s-mcp-server is in your PATH
• Kubernetes connection issues: Verify your KUBECONFIG path is correct
• Permission errors: Ensure your kubeconfig has the necessary RBAC permissions
• Extension not loading: Restart VS Code after configuration changes