product: maestro audience: test-developer, operator, ai-assistant authority: normative

MCP Server — Setup and Configuration

The Maestro MCP server runs as a Docker container inside the station stack. It exposes an HTTP MCP endpoint so any AI assistant can connect with a single URL — no local .NET SDK or dotnet run command needed.

1. Start the full stack

.\scripts\start-fullstack.ps1

The MCP server starts automatically as part of the stack. It is healthy when http://localhost:7004/health returns 200.

2. Configuration

All configuration is set in docker-compose.yml via environment variables. Nothing needs to change for a default single-station deployment.

To set the station name, pass it to the start script:

.\scripts\start-fullstack.ps1 -StationName "ST-01"

3. Connecting to an AI assistant

Connect any MCP-compatible client to:

http://<station-ip>:7004/mcp

Use localhost for local development. Use the station machine's IP address to connect from another machine on the network.

.mcp.json (repo root — used automatically by VS Code and Copilot)

{
  "mcpServers": {
    "maestro-station": {
      "type": "http",
      "url": "http://localhost:7004/mcp"
    }
  }
}

Client-specific instructions

GitHub Copilot (VS Code / Visual Studio)

Copilot reads .mcp.json from the workspace root automatically. Open the repo folder, enable MCP in Copilot settings, and start a new Copilot Chat. The maestro-station server appears in the tool list.

Claude Desktop

1. Open Settings → Developer → Edit Config and add the mcpServers block:

{
  "mcpServers": {
    "maestro-st01": {
      "type": "http",
      "url": "http://192.168.1.50:7004/mcp"
    }
  }
}

2. Restart Claude Desktop.
3. Ask: "Call get_station_info" to verify.

Cursor

Add an entry in Cursor MCP settings pointing to http://<station-ip>:7004/mcp.

mcp add (CLI)

mcp add maestro-st01 http://192.168.1.50:7004/mcp

4. Verifying the connection

Ask the AI assistant:

"Call get_station_info and tell me what station you are connected to."

A successful response looks like:

{
  "station": "ST-01",
  "stationId": "ST-01",
  "version": {
    "apiVersion": "1.5.1",
    "runners": [
      { "identifier": "dotnet", "runnerType": "DotNet", "version": "1.0.0", "healthy": true }
    ]
  }
}

If the tool call fails, check:

1. The stack is running — curl http://localhost:7004/health returns 200.
2. The station firewall allows inbound TCP on port 7004 if connecting remotely.
3. The AI client supports type: http MCP transport (all modern clients do).

5. Multi-station deployments

Each station runs its own independent stack. Add one entry per station in your client config to connect an AI assistant to multiple stations simultaneously:

{
  "mcpServers": {
    "station-01": { "type": "http", "url": "http://192.168.1.50:7004/mcp" },
    "station-02": { "type": "http", "url": "http://192.168.1.51:7004/mcp" },
    "station-03": { "type": "http", "url": "http://192.168.1.52:7004/mcp" }
  }
}

Each server instance returns its own station label in every tool response so the AI can attribute results correctly.