MCP

Manage Kubeara from Claude, Cursor, ChatGPT, and VS Code with Model Context Protocol.

Kubeara speaks Model Context Protocol (MCP) natively. AI clients use one protocol and one permission model instead of a new SSH key per assistant.

MCP is often described as USB-C for AI: a common way for tools to call APIs, databases, and platforms. Kubeara exposes your servers, services, and deploys through that protocol.

What you can do

From a connected client you can typically:

Ask which servers are connected
Deploy or inspect catalog services
Read recent deploy logs
Troubleshoot without leaving the editor or chat

You still need a Kubeara account (Cloud) or a self-hosted panel, plus an API token or OAuth where the client supports it.

Official guides

Step-by-step client setup lives on the marketing MCP section (screenshots and copy-paste configs):

What MCP is

Architecture: host, client, server, and how a request flows.

Claude Desktop

Connect Claude to Kubeara.

Cursor

Query infrastructure from the editor.

ChatGPT

OAuth-style access from ChatGPT.

VS Code

Copilot + MCP in VS Code.

Endpoint

Cloud MCP HTTP endpoint:

https://api.kubeara.dev/api/mcp

Authenticate with a bearer token from your Kubeara account. Self-hosted panels expose MCP on your panel origin — use the URL shown in the dashboard.

Example Cursor / Claude-style config:

{
  "mcpServers": {
    "kubeara": {
      "url": "https://api.kubeara.dev/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN_HERE"
      }
    }
  }
}

Never paste a production token into a public gist or GitHub issue. Rotate the token if it leaks.

Permissions

MCP uses the same authorization as your user: it cannot do more than your Kubeara role allows. That is intentional — assistants should not become a backdoor around RBAC.

Security — why MCP is not “a new SSH key per tool”