kai

Kai is a Kubernetes MCP server that acts as a bridge between large language models (LLMs) and your Kubernetes cluster. It exposes a set of natural language tools to manage core Kubernetes resources such as pods, deployments, jobs, cron jobs, services, ingresses, config maps, secrets, and namespaces. It also provides context management for switching and listing contexts, and utilities like port forwarding to pods and services. With Kai, you can issue natural language commands and have Kai translate them into Kubernetes API operations, returning structured results that your LLM can reason about or present to you directly. The server connects to your current kubectl context by default, so you can leverage your existing kubeconfig and cluster access credentials.

To use Kai, install the binary, run it, and configure your MCP consumer (Claude Desktop, Cursor, Continue, or web clients) to point at the Kai command. You can operate in standard I/O mode or SSE mode for web-based clients. Typical tasks include listing resources, creating resources (pods, deployments, services, cron jobs), retrieving details, updating configurations, and streaming logs from pods. Kai also supports common Kubernetes patterns like port-forwarding to enable local access to services running inside the cluster.

Prerequisites:

• A Kubernetes cluster accessible via kubectl (kubectl configured with a kubeconfig that can reach your cluster).
• Go installed on your machine (to build or install Kai from source, if needed).
• Optional: access tokens or certificates if your cluster requires them.

Installation steps (recommended):

1. Install Kai from the Go package repository:

go install github.com/basebandit/kai/cmd/kai@latest

2. Ensure the kai binary is in your PATH. You can verify installation with:

kai --version

3. Start Kai (default transport is stdio; for web clients you can use SSE):

kai

4. Configure your MCP client ( Claude Desktop, Cursor, Continue, or web client) to point to the Kai command as described in the README under Configuration. For example, set the command to /path/to/kai in your Claude Desktop or Cursor settings.

Tips and notes:

• Kai relies on your existing kubectl context. If you need to target a different cluster, use a kubeconfig path or context via Kai's options (as documented in the README).
• Transport modes: use stdio for local CLI-style interactions or -transport=sse for web-based clients that connect over SSE to http://localhost:8080/sse.
• Default kubeconfig path is ~/.kube/config, but you can override it with -kubeconfig and -context when launching Kai.
• Logs are emitted in structured JSON by default, which makes them easy to parse in your tooling.
• Not all Kubernetes resources are currently exposed in Kai (per the Features list in the README). The core workloads (Pods, Deployments, Jobs, CronJobs), networking (Services, Ingress), configuration (ConfigMaps, Secrets, Namespaces), context management, and port forwarding are supported. If you rely on cluster health metrics or CRDs, you may need to monitor availability or extend Kai accordingly.