概览
What is MKP - Model Kontext Protocol Server for Kubernetes?
MKP is a Model Context Protocol (MCP) server for Kubernetes that enables LLM-powered applications to list, get, apply, and execute commands on Kubernetes resources. It communicates directly with the Kubernetes API server using API Machinery’s unstructured client, supports all resource types including CRDs, and requires no external CLI tools like kubectl or helm.
How to use MKP - Model Kontext Protocol Server for Kubernetes?
Prerequisites: Go 1.24, a Kubernetes cluster with kubeconfig, and Task. Clone the repository, run task install to install dependencies, then task build. Start the server with task run (default kubeconfig) or KUBECONFIG=/path task run-with-kubeconfig. To set a custom port, use MCP_PORT=9091 task run. MKP can also be deployed via ToolHive.
Key features of MKP - Model Kontext Protocol Server for Kubernetes
- List Kubernetes resources, both clustered and namespaced
- Get resources and their subresources (status, logs, etc.)
- Apply (create/update) clustered and namespaced resources
- Execute commands in pods with configurable timeout
- Universal resource support via the unstructured client
- Built-in rate limiting and annotation filtering
Use cases of MKP - Model Kontext Protocol Server for Kubernetes
- LLM-powered applications managing Kubernetes resources
- Automated cluster inspection and troubleshooting
- Executing commands in pods for debugging
- Applying Kubernetes manifests from AI assistants
- Querying resource status, logs, and configuration
FAQ from MKP - Model Kontext Protocol Server for Kubernetes
What are the prerequisites for running MKP?
Go 1.24, a Kubernetes cluster with a valid kubeconfig, and Task (taskfile.dev) for running build and run commands.
How does MKP communicate with Kubernetes?
It uses Kubernetes API Machinery directly – no kubectl, helm, or other CLI tools are required.
Does MKP support Custom Resource Definitions (CRDs)?
Yes. The unstructured client automatically handles any resource type, including CRDs, without hardcoded schemas.
Is there rate limiting?
Yes, MKP includes built-in rate limiting to protect against excessive API calls.
What is the default timeout for pod exec commands?
The default timeout is 15 seconds, with a maximum of 60 seconds. Commands exceeding the timeout are terminated and return an error.
