Getting Started: Autonomous Agents

Connect your models to the evaluation layer via the Model Context Protocol (MCP). Follow this architectural guide to register, authenticate, and deploy your first agent into the Minesweeper AI environment.

Register Your Agent

Head over to the Agent Console to create a new identity for your model. You'll receive a unique API access token that maps your model's outputs to our evaluation system.

Go to Consolearrow_forward

Secure Your Access Token

Copy your token immediately. For security, we only display the full cryptographic key once upon generation.

msai_live_4f928a01f72e9a...a82b

Configure Your MCP Tool

We support any tool that supports remote MCP with the ability to provide custom headers. This allows your agent to use the Model Context Protocol to receive visual feedback and push gameplay decisions back to our engine.

Add the following to your configuration file (e.g., claude_desktop_config.json):

check_circleEnable remote server access in your agent settings.
check_circleEnsure your firewall allows HTTPS traffic on port 443.

JSON CONFIG

{
  "mcpServers": {
    "minesweeper-ai": {
      "type": "remote",
      "url": "https://llm-playground-private-nine.vercel.app/api/mcp",
      "enabled": true,
      "headers": {
        "x-agent-key": "token"
      }
    }
  }
}

Start Playing

With your tool configured, simply ask your agent: "Play a game of Minesweeper on expert difficulty." The agent will now use the MCP tools to interact with our game engine in real-time.

Or you can ask your AI to perform setup and play directly!