A native MCP Server implementation for Kubernetes and OpenShift, enabling AI Agents to directly manage cluster resources with CRUD operations, Pod management, Helm/Tekton/KubeVirt toolsets, read-only mode, and multi-cluster support.
kubernetes-mcp-server is an open-source project maintained by the Red Hat containers organization, providing a native Model Context Protocol (MCP) Server implementation for Kubernetes and OpenShift. Unlike simple kubectl/helm CLI wrappers, it directly calls the Kubernetes API Server via Go, distributed as a single binary with high performance and low latency.
Core capabilities cover generic Kubernetes resource CRUD, full Pod lifecycle management (logs, Exec, Top, run image), Namespace/Events viewing, Helm Chart management, and Tekton Pipeline operations. Extended toolsets (KubeVirt, Kiali, KCP) are enabled on demand.
Security controls include
--read-onlymode,--disable-destructivemode, TOML-configurable denied resource rules (e.g., Secret), and OAuth/OIDC authentication (Keycloak, Microsoft Entra ID) in HTTP mode.Flexible deployment: native binary, npm package, Python package, Docker image, and Helm Chart (in-cluster). MCP transport supports both stdio and HTTP (Streamable HTTP + SSE). Configuration via CLI args, TOML files (with SIGHUP hot-reload), or environment variables.
Typical use cases: natural language-driven cluster management via Claude Desktop/Cursor/Copilot, automated fault diagnosis and repair, Vibe Coding to deployment workflows, read-only cluster inspection, and multi-cluster unified management. Optional OpenTelemetry distributed tracing and metrics are supported for observability.
Note: The exact URLs for the npm package, Python package (PyPI), and Docker image are not explicitly provided in the README and remain unconfirmed.