Connect an AI Assistant (MCP)

The MCP server lets an AI assistant such as Claude find a user's devices, read their state, change it, and manage their schedules, using the Model Context Protocol. Use this page to connect an assistant to your deployment; the assistant can then act only within the permissions the signed-in user already has.

The MCP server is deployed with the rest of the backend — there is nothing to enable.

What the assistant can do

Once connected, ask in plain words:

• "Which lights are on in the kitchen?"
• "Set the bedroom fan to 60%."
• "Turn the porch light on every weekday at 7 pm."

The assistant answers these with five tools:

Tool	What it does
list_devices	Finds devices by name, room or type, and reads their current state
list_groups	Lists the user's groups (homes) and the subgroups (rooms) inside them
set_params	Changes what a device is doing right now, such as power or brightness
list_schedules	Lists the schedules stored on one device
set_schedule	Creates, changes, removes, enables or disables a schedule on one device

Every tool's inputs and outputs are listed in the MCP Tools reference.

The assistant cannot add, remove or rename devices, change groups, or report past usage.

1. Find your MCP URL

1. Open your deployment's rmng-client-outputs.json — see Deployment Details for where to find it.

2. Copy the value of McpHttpApiUrl under rmng-core:

   rmng-client-outputs.json

   {
     "rmng-core": {
       "McpHttpApiUrl": "https://<api-id>.execute-api.<region>.amazonaws.com/"
     }
   }

3. Append v1/mcp. The result is your MCP URL:

   MCP URL

   https://<api-id>.execute-api.<region>.amazonaws.com/v1/mcp

2. Add it to your assistant

Pick your assistant. Every option ends the same way: a browser window opens, and you sign in with your ESP RainMaker Neo user.

Claude Code

Claude Code is Anthropic's command-line coding assistant; its VS Code and Cursor extensions share the same settings.

1. Add the server:

   claude mcp add --transport http --scope user rainmaker-neo <mcp-url>

   --scope user makes it available in every folder. Use --scope project instead to save it in a .mcp.json file your team can commit.

2. Start claude and type /mcp.

3. Select rainmaker-neo, then Authenticate, and sign in. The server shows as connected in /mcp.

Claude app

For the Claude desktop app and claude.ai:

1. Open Settings → Connectors → Add custom connector.
2. Paste the MCP URL and save.
3. Select Connect and sign in. The connector shows as connected.

VS Code

For GitHub Copilot in VS Code, add the server to .vscode/mcp.json:

.vscode/mcp.json

{
  "servers": {
    "rainmaker-neo": { "type": "http", "url": "<mcp-url>" }
  }
}

Cursor

1. Open Settings (gear icon, top right).

2. Go to Tools & Integrations → MCP Tools.

3. Select New MCP Server. Cursor opens ~/.cursor/mcp.json.

4. Add the server and save:

   ~/.cursor/mcp.json

   {
     "mcpServers": {
       "rainmaker-neo": { "url": "<mcp-url>" }
     }
   }

5. Select Connect next to rainmaker-neo and sign in.

Not yet tested against ESP RainMaker Neo sign-in.

Other

Any assistant that supports remote MCP servers over HTTP can use the MCP URL directly. For an assistant that can only launch local servers, bridge through mcp-remote:

{
  "mcpServers": {
    "rainmaker-neo": { "command": "npx", "args": ["mcp-remote", "<mcp-url>"] }
  }
}

The assistant must support CIMD sign-in

The server identifies assistants by a Client ID Metadata Document (CIMD): an HTTPS link to a small file describing the assistant. It does not support dynamic client registration, so an assistant that can only register itself that way will fail at sign-in. If your deployment sets MCP_CIMD_ALLOWED_HOSTS, the assistant's CIMD host must also be on that list.

Related

• MCP specification — how the tools and sign-in work.
• MCP Tools reference — every tool and its inputs.
• Voice Assistants — link Alexa, Google Home or SmartThings.
• Deployment — deploy the backend that hosts the server.