Kubernetes AI Management

You run an AI-powered MCP server that analyzes your Kubernetes cluster, provides real-time diagnostics, resource insights, and smart log analysis through an MCP client. It lets you ask natural questions about cluster health, workloads, networking, storage, and deployments, and receive actionable guidance without memorizing commands.

How to use

You interact with the MCP client to query and control your Kubernetes environment. Ask natural questions like “What’s the health of my cluster?”, “Show me pods in default namespace with high restart counts”, or “Upgrade a Helm release.” The MCP server processes your requests, analyzes cluster state, and returns actionable steps, dashboards, or concrete commands you can run in your terminal or CI/CD workflow. You can run in two modes: MCP server mode (self-hosted server) or Agent mode (Rest API) depending on your integration needs.

How to install

Prerequisites you need before running the MCP server or agent: a Java Development Kit (JDK) 17 or later, Maven 3.8 or later, and a configured Kubernetes cluster with kubeconfig available at ~/.kube/config.

Build and run the MCP server or agent locally as shown in the commands below.

# Build all modules
mvn clean package

# Run the MCP server
java -jar mcp-server/target/mcp-server-1.0-SNAPSHOT.jar

# Alternatively, run the agent directly
java -jar agent/target/agent-*-fat.jar

Additional sections

Configuration and usage notes help you tailor the MCP experience to your cluster. If you prefer a web-based or REST API-driven workflow, you can use the Agent mode with its Rest API, or run the MCP server for a full-hosted experience.

Security considerations: ensure access to the MCP server and agent is restricted to trusted clients. Use TLS/HTTPS for remote access if exposing the API, and apply RBAC controls in your Kubernetes cluster to limit what the MCP components can read or modify.

Troubleshooting tips: verify your kubeconfig is correct, confirm the MCP server and agent JARs are built successfully, and check Java runtime logs for initialization errors. If a diagnostic command returns unexpected results, re-run the relevant module build and restart the corresponding service.

Available tools

cluster_health

Analyzes overall cluster health, detects unhealthy components, and provides recommended remediation steps.

network_analysis

Examines network topology, endpoints, ingress/egress rules, and service exposure to identify connectivity or exposure issues.

storage_management

Monitors persistent volumes, claims, storage classes, and capacity to forecast shortages and optimize usage.

job_analysis

Inspects Jobs and CronJobs, reports failures, history, and upcoming schedules.

helm_release_management

Manages Helm releases, including listing, upgrading, rolling back, and inspecting values and histories.