MCPServer stays Unhealthy with uvx/npx unless image is explicitly set #76
Description
Summary
Creating an MCPServer that uses npx (and likely uvx) results in the Deployment being created but the MCPServer never becomes Healthy unless spec.deployment.image is explicitly provided. This suggests no default runtime image is selected for uvx/npx commands.
Repro
- Apply:
apiVersion: kagent.dev/v1alpha1
kind: MCPServer
metadata:
name: everything
namespace: test
spec:
deployment:
cmd: npx
args: [modelcontextprotocol/server-everything]
# image omitted on purpose
port: 3000
stdioTransport:
transportType: stdio
- Verify status:
kubectl get mcps -n test
NAME HEALTH AGE
everything Unhealthy 2m31s
kubectl get deploy,pods -n test -l app=everything
NAME READY UP-TO-DATE AVAILABLE AGE
deployment/everything 1/1 1 1 2m31s
NAME READY STATUS RESTARTS AGE
pod/everything-xxxxx-xxxxx 1/1 Running 0 2m31s
- Observe: Deployment/Pod exist; MCPServer status stays Unhealthy.
Workaround
- Set spec.deployment.image: node:24-alpine3.21.
After reconcile:
kubectl get mcps -n test
NAME HEALTH AGE
everything Healthy 3m10s
Expected
- Auto-select a suitable default image for uvx/npx or fail fast with a clear validation error.
Actual
- No default image; resource remains Unhealthy without a clear hint.
Metadata
Metadata
Assignees
Labels
Type
Projects
Status