kind

HPC MCP Server Example

Run the mcpserver in Kubernetes (or OpenShift) with basic manifests

Prerequisites

• Docker
• kind (Kubernetes in Docker)
• kubectl
• Python 3.11+ (for local testing)

1. Create a Kind Cluster

Start by creating a local Kubernetes cluster (with kind):

kind create cluster

2. Build and Load the Image

Build the container image locally and load it into the kind nodes so you don't have to push to a remote registry. You can also pull and load.

# Build the image
docker build -t ghcr.io/converged-computing/mcp-server:latest .

# OR pull
docker pull ghcr.io/converged-computing/mcp-server:latest

# Load into kind
kind load docker-image ghcr.io/converged-computing/mcp-server:latest

3. Deploy the Server

The configuration and the custom tools are managed via a Kubernetes ConfigMap. This contains our mcpserver.yaml (the configuration) and echo.py (the tool code).

kubectl apply -f config-map.yaml

This launches the pod, mounts the configuration, and exposes it via a Service.

kubectl apply -f deployment.yaml
kubectl apply -f service.yaml

Wait for the pod to reach the Running state:

kubectl get pods

See the server running:

kubectl logs mcp-server-5bbdcbbbdf-2sccc -f

Expose the service to your local machine:

kubectl port-forward svc/mcp-server-service 8080:80

Check health:

$ curl -s http://localhost:8080/health  | jq
{
  "status": 200,
  "message": "OK"
}

Ask for pancakes (you need fastmcp installed for this).

$ python3 get_pancakes.py 
  ⭐ Discovered tool: pancakes_tool
  ⭐ Discovered tool: simple_echo

CallToolResult(content=[TextContent(type='text', text='Pancakes for Vanessa 🥞', annotations=None, meta=None)], structured_content={'result': 'Pancakes for Vanessa 🥞'}, meta=None, data='Pancakes for Vanessa 🥞', is_error=False)

Note that we can also run the server in stdio mode and then echo json RPC to it, but nah, don't really want to do that.

Clean Up

To delete the cluster and start over:

kind delete cluster