Kubectl MCP Server

What rohitg00/kubectl-mcp-server MCP server does

The rohitg00/kubectl-mcp-server MCP server connects MCP-compatible AI assistants to Kubernetes clusters. It is designed for conversational operations such as investigating crashed pods, examining logs and events, deploying applications, tracing service connectivity, reviewing resource usage, and auditing cluster security.

The README lists 253 tools, eight workflow prompts, and eight data resources. The server supports both read-oriented investigation and management tasks, so it should be treated as an interface to live cluster operations rather than only a documentation or query tool. Kubernetes access remains controlled by the cluster configuration and permissions available to the process running the server.

How it works

The server runs locally or in a container and communicates with an MCP client. The documented clients include Claude Desktop, Cursor, Windsurf, Copilot, and other AI tools. Requests written in natural language are translated into Kubernetes-related operations exposed through MCP.

A configured kubectl installation and access to a Kubernetes cluster are prerequisites. The README also documents stdio operation and an HTTP mode exposed through Docker with the SSE transport option. Optional features include interactive HTML dashboards with live metrics, browser automation for web-based tools, and an in-cluster deployment approach called kMCP.

Setup and configuration

The rohitg00/kubectl-mcp-server MCP server can be started without a local package installation using Node.js and npx:

npx -y kubectl-mcp-server

It can also be installed from PyPI with Python 3.9 or newer:

pip install kubectl-mcp-server
pip install "kubectl-mcp-server[ui]"

The optional ui extra is documented for interactive dashboards. Docker images are published through Docker Hub and GitHub Container Registry. The README shows mounting $HOME/.kube into the container, read-only, so the process can use the local Kubernetes configuration.

Before connecting an assistant, the project provides CLI checks for cluster information, server details, available tools, and a sample tool call. The exact assistant configuration depends on the client; the README includes a Claude Desktop setup section and demonstrations for Claude Desktop, Cursor, and Windsurf.

Tools and capabilities

Documented use areas include:

• Pod, log, event, and resource analysis for troubleshooting
• Application deployment, including replica, service, storage, and health-check configuration
• Resource waste and cost-optimization analysis
• Network and service-chain diagnostics, including DNS-related checks
• RBAC analysis, secret scanning, and pod security policy review
• Helm chart management
• Interactive dashboards and visual event or resource information
• Browser automation for web-based Kubernetes tools

The server also advertises non-destructive mode, secret masking, RBAC validation, OAuth 2.1 authentication for enterprise use, multi-cluster support, and agent skills for AI coding agents. The README does not provide the detailed behavior or configuration for each of these features in the supplied material.

Limitations and notes

The server requires an accessible Kubernetes cluster and a working kubectl configuration; installing the MCP package alone does not provide a cluster. Actions that change infrastructure depend on the permissions granted to the configured Kubernetes identity. Dashboard functionality requires the optional UI installation when using the documented Python setup.

The README identifies the project as MIT-licensed and lists Python 3.9+ and Node.js 14+ as supported prerequisites for the corresponding installation paths. Review client-specific configuration and cluster permissions before enabling write operations in a production environment.