harvester

Harvester MCP Server enables AI assistants like Claude Desktop and Cursor to interact with Harvester clusters via the MCP protocol. It translates natural language commands into Kubernetes/Harvester API operations, formats results into human-readable text, and returns them to the AI client for presentation. The server exposes tools for both Kubernetes core resources (Pods, Deployments, Services, Namespaces, Nodes, etc.) and Harvester-specific resources (Virtual Machines, Images, Volumes, Networks), with standardized formatters to produce concise lists and detailed views optimized for LLM consumption.

To use the server, run it with access to a kubeconfig for your Harvester cluster (or run inside a cluster to use in-cluster configuration). Then configure your MCP client (e.g., Claude Desktop) to point at the Harvester MCP Server endpoint. Once connected, you can ask high-level questions like: “List all virtual machines in the default namespace” or “Show details for the pod named rancher-789c976c6-xbvmd in cattle-system.” The MCP server will handle request parsing, resource handling, formatting, and response delivery, providing human-friendly outputs suitable for AI assistants to present to users.

Prerequisites:

• Go 1.23+ installed on your machine
• Access to a Harvester cluster with a kubeconfig file (or run inside a Kubernetes cluster with proper permissions)

From Source

1. Clone the repository:

git clone https://github.com/starbops/harvester-mcp-server.git
cd harvester-mcp-server

2. Build the server binary:

make build

3. Run the server (requires kubeconfig or in-cluster config):

./bin/harvester-mcp-server

Using Go install

go install github.com/starbops/harvester-mcp-server/cmd/harvester-mcp-server@latest

3. Run the installed binary (if installed via Go install):

harvester-mcp-server

Optional with kubeconfig path or log level:

harvester-mcp-server --kubeconfig=/path/to/kubeconfig.yaml --log-level=debug

Tips and considerations:

• Ensure your kubeconfig has sufficient privileges to read the resources you want to query (rbac considerations).
• The server automatically discovers kubeconfig in this order: in-cluster config, --kubeconfig flag, KUBECONFIG env var, default ~/.kube/config.
• If you plan to run inside a Kubernetes cluster, consider running with appropriate service account permissions and configuring network access for your MCP clients.
• When using Claude Desktop or similar tools, add the Harvester MCP server under the mcpServers section with the correct command path and, optionally, kubeconfig and log level.
• If you add new resources, register their tooling in the MCP server (types.go mappings, formatters, and server registration).
• For debugging, run with --log-level=debug and inspect the formatted outputs for any adjustments needed in formatter logic.