Quick Start

Two ways to a running agent: the tiny CLI against your own cluster (the usual way), or the hosted MCP endpoint if you'd rather not install anything.

With tiny — your cluster, one binary

brew install tiny-systems/tap/tiny   # or: go install github.com/tiny-systems/tiny@latest
tiny up                              # CRDs, broker, operator, core modules — any kind/k3s/cloud cluster
tiny mcp                             # exposes the MCP server for your editor

Connect Claude Code once:

claude mcp add --transport http tiny http://localhost:7776/mcp

Then prompt your agent:

Show me what's installed, then build an agent that ticks every 5 seconds and logs to a debug node.

The MCP tools do the work — list_modules, get_component_info, build_flow — and you can watch it in your cluster:

kubectl get tinynodes -n tinysystems
kubectl get tinyflows -n tinysystems

No account, no hosted anything: those are your pods in your namespace. Get tiny has all install options.

From your editor: hosted MCP

Prefer zero install? Drive agents from Claude Code, Cursor, or any MCP client against a hosted workspace on your own connected cluster.

  1. Sign in at tinysystems.io and connect a Kubernetes cluster.
  2. Add the MCP endpoint to your client:
{
  "mcpServers": {
    "tinysystems": {
      "url": "https://mcp.tinysystems.io/mcp",
      "headers": { "Authorization": "Bearer <your-workspace-key>" }
    }
  }
}

Full walkthrough: MCP Server Setup.

What's next