Kruize MCP Server

A cloud-native Model Context Protocol (MCP) server that bridges AI assistants and the Kruize resource optimization engine. It exposes Kubernetes workload recommendations — CPU, memory, and idle workload detection — as structured tools that any MCP-compatible AI client can invoke.

AI Client (Claude, Bob, etc.)
        │  MCP over HTTP/SSE
        ▼
Kruize MCP Server  ←→  Kruize API  ←→  Kubernetes cluster

Prerequisites

Tool	Purpose
Java 21+	Build and run the server (Local JAR only)
`kubectl` or `oc`	Deploy to Kubernetes / OpenShift
Kruize running	The backend recommendations engine

Need Kruize? → git clone https://github.com/kruize/kruize-demos.git and run local_monitoring_demo.sh. See the Deployment Guide for details.

Quick Start

Local JAR (fastest)

git clone https://github.com/kruize/kruize-mcp-server.git
cd kruize-mcp-server
./mvnw clean package -DskipTests
KRUIZE_URL=http://<kruize-host>:<port> java -jar target/kruize-mcp-server-0.0.1-runner.jar

MCP endpoint: http://localhost:8080/mcp

Minikube

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

# 2. Patch the manifest, then deploy
sed -i.bak "s|http://<minikube-ip>:<kruize-port>|$KRUIZE_URL|g" manifests/kruize-mcp-server-minikube.yaml
kubectl apply -f manifests/kruize-mcp-server-minikube.yaml
kubectl wait --for=condition=ready pod -l app=kruize-mcp-server -n monitoring --timeout=120s

# 3. Port-forward
kubectl port-forward -n monitoring svc/kruize-mcp-server-service 8082:8082

MCP endpoint: http://localhost:8082/mcp

OpenShift

git clone https://github.com/kruize/kruize-mcp-server.git
cd kruize-mcp-server
oc apply -f manifests/kruize-mcp-server-openshift.yaml -n openshift-tuning

# Get the route URL
MCP_ROUTE=$(oc get route kruize-mcp-server-service -n openshift-tuning --template='http://{{ .spec.host }}')
echo "MCP endpoint: $MCP_ROUTE/mcp"

MCP endpoint: http://<route-host>/mcp

For full step-by-step instructions and troubleshooting see the Deployment Guide.

Connect an AI Client

Once the server is running, connect your AI client to the MCP endpoint.

MCP Inspector (browser-based testing)

npx @modelcontextprotocol/inspector <mcp-endpoint>

Claude Desktop

Edit your Claude Desktop config file (claude_desktop_config.json):

{
  "mcpServers": {
    "kruize": {
      "url": "http://localhost:8080/mcp",
      "type": "streamable-http"
    }
  }
}

Claude Code

claude mcp add --transport streamable-http kruize http://localhost:8080/mcp

Bob

Add to your ~/.bob/settings/mcp.json:

{
  "mcpServers": {
    "kruize": {
      "url": "http://localhost:8080/mcp",
      "type": "streamable-http"
    }
  }
}

Replace http://localhost:8080/mcp with your actual MCP endpoint (e.g. the OpenShift route URL or http://localhost:8082/mcp for Minikube). See the Deployment Guide for all client options and multi-endpoint setup.

Available Tools

Tool	Description
`listAllExperiments`	List all Kruize experiments
`listAllRecommendations`	CPU/memory recommendations for all containers
`getCostOptimizedRecommendations`	Cost-optimized sizing, optionally filtered by namespace
`getPerformanceOptimizedRecommendations`	Performance-optimized sizing, optionally filtered by namespace
`getIdleWorkloads`	Workloads with near-zero CPU usage; optionally include recommendations

Example prompts:

"Which workloads in the production namespace are over-provisioned for CPU?"
"Show me idle workloads with cost-saving recommendations."
"What are the performance recommendations for the auth service?"

Building and Pushing the Container Image

The scripts/build_and_push.sh script handles multi-arch image builds and pushes using either Docker (buildx) or Podman. The Dockerfile uses a multi-stage build — compilation happens inside the container, so no local JDK or Maven installation is required.

Options

Flag	Default	Description
`-i IMAGE`	—	Full image reference (e.g. `quay.io/user/kruize-mcp-server:0.0.1`); overrides `-r`, `-n`, `-t`
`-r REGISTRY`	`quay.io`	Container registry hostname
`-n REPO_NAME`	`kruize/kruize-mcp-server`	Repository name
`-t TAG`	`latest`	Image tag
`-l PLATFORMS`	`linux/amd64,linux/arm64`	Comma-separated target platforms
`-p PUSH`	`false`	Push image after build (`true`/`false`)

All options can also be set via environment variables (REGISTRY, REPO_NAME, IMAGE_TAG, PLATFORMS, PUSH_IMAGE).

Examples

# Build and push using a full image reference (simplest form)
./scripts/build_and_push.sh -i quay.io/user/kruize-mcp-server:0.0.1 -p true

# Build only (no push)
./scripts/build_and_push.sh -i quay.io/user/kruize-mcp-server:0.0.1

# Build for a single platform
./scripts/build_and_push.sh -i quay.io/user/kruize-mcp-server:dev -l linux/amd64

# Build and push using individual flags
./scripts/build_and_push.sh -r docker.io -n myorg/kruize-mcp-server -t latest -p true

Documentation

Doc	Description
Deployment Guide	Full setup, client config, and troubleshooting for all environments
MCP Tools Reference	Tool parameters, example prompts, and sample output
Health Check API	Health endpoint reference

License

Apache 2.0

FilesExpand file tree

README.md

Latest commit

History