examples

Claude Desktop Configuration Examples

Two ready-to-paste snippets for Claude Desktop's MCP config.

Files

File	What it enables
claude_desktop_config.read_only.json	All 13 read tools. Safe default — no way to change cluster state.
claude_desktop_config.with_writes.json	All 13 read tools plus the 3 write tools (scale_deployment, restart_deployment, delete_pod), restricted to the dev and staging namespaces. Each write defaults to dry_run=True; the LLM must explicitly pass dry_run=False to apply.

Placeholder to replace

Both files contain /Users/yourname/.kube/config — replace with your actual kubeconfig path before saving. On Windows that's typically C:\\Users\\yourname\\.kube\\config (note the doubled backslashes — JSON requires them).

Where Claude Desktop's config lives

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Claude Desktop is officially macOS + Windows only. The MCP server itself runs on any platform Python supports (including Linux) — useful with Claude Code or other MCP clients that don't ship a desktop GUI.

Merging with existing servers

Claude Desktop wants ONE mcpServers object containing every server. If you already have other MCP servers configured, don't replace the whole file — add "kubernetes" as a new key under your existing mcpServers:

{
  "mcpServers": {
    "your-other-server": { ... },
    "kubernetes": {
      "command": "k8s-mcp-server",
      "args": ["--kubeconfig", "/Users/yourname/.kube/config"]
    }
  }
}

After editing, fully quit and reopen Claude Desktop (not just close the window — use Cmd-Q on macOS).

Security warning — --enable-writes

The with_writes.json example enables destructive operations: scale, restart, and delete. Even with all the safeguards (dry-run by default, namespace allowlist, audit logging), think carefully before pointing this at any cluster you care about.

The minimum-surprise recipe:

1. Always use --namespaces to restrict the blast radius. The example uses dev,staging.
2. Never enable writes against a production cluster from a personal laptop. Run a separate, scoped instance from an audited environment for any production-write workflow.
3. Read docs/SECURITY.md for the full defense-in-depth model — the five layers, the audit logger contract, and the secret-handling rules.

If you're unsure whether to enable writes, don't. Start with the read-only example and graduate later.