Kruize

A cloud-native MCP server designed to act as a bridge between AI models/Agentic workflows and the Kruize recommendations engine. Kruize MCP server exposes Kubernetes resource recommendation data, including idle workload detection, as simple, actionable tools for AI models.

Quick Start

Choose your platform:

• OpenShift - Deploy on OpenShift clusters
• Minikube - Deploy on Minikube for local development

---

OpenShift

Deploy

# 1. Clone and build
git clone https://github.com:kruize/kruize-mcp-server.git
cd kruize-mcp-server
./mvnw install

# 2. Deploy Kruize
./local_monitoring_demo.sh -c openshift -e container

# 3. Build and push image
docker build -t <registry>/<username>/kruize-mcp-server:<tag> .
docker push <registry>/<username>/kruize-mcp-server:<tag>

# 4. Deploy MCP server (in openshift-tuning namespace)
oc apply -f manifests/kruize-mcp-server-openshift.yaml -n openshift-tuning
oc expose service kruize-mcp-server-service -n openshift-tuning

# 5. Get URL and connect Inspector
oc get route kruize-mcp-server-service -n openshift-tuning --template='{{ .spec.host }}'
npx @modelcontextprotocol/inspector http://<route-url>/mcp/

Deployment Location: openshift-tuning namespace (same as Kruize)

---

Minikube

Deploy

# 1. Clone
git clone https://github.com:kruize/kruize-mcp-server.git
cd kruize-mcp-server

# 2. Deploy Kruize (sets up Minikube + Prometheus)
git clone https://github.com/kruize/kruize-demos.git
cd kruize-demos/monitoring/local_monitoring
./local_monitoring_demo.sh -c minikube -f -e container

# 3. Get Kruize connection details
# Get Kruize URL (Minikube IP + NodePort)
KRUIZE_URL=$(echo "http://$(minikube ip):$(kubectl get svc kruize -n monitoring -o jsonpath='{.spec.ports[0].nodePort}')")
echo "Kruize URL: $KRUIZE_URL"

# 4. Deploy MCP server
cd kruize-mcp-server
# Update the KRUIZE_URL in the manifest file with the value from above
# Edit manifests/kruize-mcp-server-minikube.yaml and replace <minikube-ip>:<kruize-port> with the actual URL
kubectl apply -f manifests/kruize-mcp-server-minikube.yaml
kubectl wait --for=condition=ready pod -l app=kruize-mcp-server -n monitoring --timeout=120s

# 5. Port forward and connect Inspector
kubectl port-forward -n monitoring service/kruize-mcp-server-service 8082:8082
npx @modelcontextprotocol/inspector http://localhost:8082/mcp/

Note: Kruize MCP server uses port 8082 (Kruize uses 8080/8081)

---

Important: After connecting with Inspector tool, verify the URL matches your deployment in the Inspector tool UI:

• OpenShift: Use the MCP server route URL from oc get route kruize-mcp-server-service -n openshift-tuning
• Minikube: Use http://localhost:8082/mcp/ (after port-forward)
• Local JAR: Use http://localhost:8080/mcp/ or http://localhost:8082/mcp/ depending on configuration

---

MCP Tools

• listAllExperiments - List all experiments
• listAllRecommendations - List resource optimization CPU/memory recommendations for all the containers
• getCostOptimizedRecommendations - Get cost-optimized CPU/memory recommendations for a container. Optionally by namespace
• getPerformanceRecommendations - Get performance-optimized CPU/memory recommendations for a container. Optionally by namespace
• getIdleWorkloads - Get idle workloads (CPU usage < 1 millicore). Optionally with cost and performance recommendations

---

Local Development

Run from JAR - OpenShift Configuration

# 1. Build the project
./mvnw clean install

# 2. Get your Kruize route URL
KRUIZE_URL=$(oc get route kruize -n openshift-tuning --template='http://{{ .spec.host }}')
echo "Kruize URL: $KRUIZE_URL"

# 3. Run the JAR file with Kruize URL (port 8080)
KRUIZE_URL=$KRUIZE_URL java -jar target/kruize-mcp-server-1.0-SNAPSHOT-runner.jar

# 4. Connect Inspector
npx @modelcontextprotocol/inspector http://localhost:8080/mcp/

Custom Kruize URL (if needed):

# Replace with your actual Kruize route URL
KRUIZE_URL=http://kruize-openshift-tuning.apps.your-cluster.com java -jar target/kruize-mcp-server-1.0-SNAPSHOT-runner.jar

Run from JAR - Minikube Configuration

# 1. Build the project
./mvnw clean install

# 2. Get Kruize URL (Minikube IP + NodePort)
KRUIZE_URL=$(echo "http://$(minikube ip):$(kubectl get svc kruize -n monitoring -o jsonpath='{.spec.ports[0].nodePort}')")
echo "Kruize URL: $KRUIZE_URL"

# 3. Run the JAR file (port 8082 to avoid conflict with Kruize on 8080/8081)
QUARKUS_HTTP_PORT=8082 KRUIZE_URL=$KRUIZE_URL java -jar target/kruize-mcp-server-1.0-SNAPSHOT-runner.jar

# 4. Connect Inspector
npx @modelcontextprotocol/inspector http://localhost:8082/mcp/

Custom Kruize URL (if needed):

# Replace with your actual Kruize URL
QUARKUS_HTTP_PORT=8082 KRUIZE_URL=http://192.168.49.2:30080 java -jar target/kruize-mcp-server-1.0-SNAPSHOT-runner.jar

---

Health Check API

The server includes comprehensive health check endpoints for monitoring and Kubernetes integration:

Quick Test

# Test health endpoints (default port)
curl http://localhost:8080/q/health        # Overall health
curl http://localhost:8080/q/health/live   # Liveness probe
curl http://localhost:8080/q/health/ready  # Readiness probe

# Test with custom URL
curl http://localhost:8082/q/health        # Custom port
curl https://kruize-mcp.example.com/q/health  # Remote server

Health Endpoints

• /q/health - Overall health status (aggregates all checks)
• /q/health/live - Liveness probe (triggers pod restart on failure)
• /q/health/ready - Readiness probe (removes pod from service on failure)

Kubernetes Integration

Health probes are automatically configured in the Kubernetes manifests:

• Liveness Probe: Checks every 30s, restarts pod on failure
• Readiness Probe: Checks every 10s, removes from service on failure

Both probes verify Kruize API connectivity to ensure the service is operational.

Documentation

• Health Check API Guide - Complete API reference