Harvester MCP Server

What is Harvester MCP Server?

Harvester MCP Server is a Go implementation of the Model Context Protocol (MCP) designed for Harvester HCI. It enables AI assistants like Claude Desktop and Cursor to perform CRUD operations on Harvester clusters through natural language, bridging AI with Kubernetes-native HCI resources.

How to use Harvester MCP Server?

Install from source (git clone && make build) or via go install. Configure with a valid kubeconfig (using --kubeconfig flag, KUBECONFIG env var, or default ~/.kube/config). For Claude Desktop, add the server binary and args to claude_desktop_config.json under mcpServers. Then ask natural-language queries about cluster resources.

Key features of Harvester MCP Server

• Kubernetes core resources: Pods (List, Get, Delete), Deployments, Services, Namespaces, Nodes (List, Get), CRDs (List)
• Harvester-specific resources: Virtual Machines (List, Get), Images, Volumes, Networks (List)
• Human-readable formatted outputs optimized for LLM consumption
• Automatic grouping of resources by namespace or status
• Concise summaries and detailed views for resource inspection
• Unified resource handler architecture for extensibility

Use cases of Harvester MCP Server

• Query cluster health: "How many nodes are in my Harvester cluster?"
• Troubleshoot workloads: "List all pods in the cattle-system namespace"
• Inspect VMs: "Show the details of the virtual machine web-server in default namespace"
• Explore services: "What services are running in the harvester-system namespace?"
• Get resource details: "List all virtual machines in the default namespace"

FAQ from Harvester MCP Server

What resources and operations are supported?

Supported operations include List, Get, and Delete for Pods; List and Get for Deployments, Services, Namespaces, Nodes, and Virtual Machines; List for CRDs, Images, Volumes, and Networks. Writing and creating resources are described in the resource handler architecture but are not yet exposed as tools.

How do I configure cluster access?

The server looks for a kubeconfig in this order: in-cluster config (if running inside a cluster), --kubeconfig flag, KUBECONFIG environment variable, or ~/.kube/config. Use harvester-mcp-server --kubeconfig=/path/to/config.yaml or set KUBECONFIG.

How do I integrate with Claude Desktop?

Add an entry to Claude Desktop’s config file (claude_desktop_config.json) under mcpServers with the command pointing to the built binary and arguments like --kubeconfig and --log-level. Restart Claude Desktop for the tools to appear.

What are the runtime requirements?

Go 1.23+ to build from source, and access to a Harvester cluster with a valid kubeconfig. No additional runtimes are needed after building the binary.

Can I change the logging verbosity?

Yes. Use the --log-level flag with values debug, info, warn, error, fatal, or panic (default info). Example: harvester-mcp-server --log-level=debug.