Open-Source kubernetes-mcp-server Lets AI Agents Run Real Clusters Without Shell Access
The containers/kubernetes-mcp-server gives AI agents a native, dependency-free interface to Kubernetes clusters, turning natural language into real cluster operations.
Takeaways
- The containers/kubernetes-mcp-server is a Go-based MCP server that gives AI agents native read-write access to Kubernetes and OpenShift clusters.
- Unlike kubectl-wrapper approaches, it talks directly to the Kubernetes API, requiring zero external dependencies or CLI tools.
- Supports pods, deployments, events, Helm, Tekton, KubeVirt, Kiali, and NetObserv via opt-in toolsets, with 32+ automated eval scenarios.
- Safety controls include
--read-only,--disable-destructive, denied-resources config, and automatic credential redaction in logs. - Works with Claude Desktop, VS Code Copilot, Cursor, and Goose CLI via npx, uvx, native binary, or Helm chart; free under Apache 2.0.
- Lands as MCP crosses 97 million installs and becomes the de facto standard for agent-to-tool communication in production infrastructure.
Managing Kubernetes has always been a human-in-the-loop job: write YAML, run kubectl, read logs, repeat. The containers/kubernetes-mcp-server project challenges that pattern directly. It exposes your entire Kubernetes or OpenShift cluster as a set of structured tools that any MCP-compatible AI agent can call, turning natural language prompts into real cluster operations with no shell wrappers involved.
Why existing integrations fall short
Most AI-to-Kubernetes integrations work by having the model generate a kubectl command and shell out to run it. That approach is fragile: it requires the CLI to be installed, adds latency, and forces the model to guess the right flags. This server takes a different path. Written in Go, it talks directly to the Kubernetes API server with no external CLI dependencies. You don't need Node or Python on the system if you're running the native binary, and there's no subprocess overhead waiting on shell commands to return.
The server ships as a single native binary for Linux, macOS, and Windows, and is also available as an npm package, a Python package, and a container image.
What an agent can actually do
Tools are organized into opt-in toolsets. Two are enabled by default: config and core. Additional toolsets unlock integrations with Helm, Tekton, KubeVirt, Kiali, and NetObserv.
With the default toolsets, an agent can:
- List, get, create, update, and delete any Kubernetes or OpenShift resource by