MCP & Agent Skills

Selora AI includes an MCP server, so an AI agent such as Claude or ChatGPT can read your home and, if you allow it, control it. You only need this to use Selora AI from an outside agent: the Selora AI chat panel and HA Assist work without it.

1. Find Your MCP Server URL

In Selora AI, open the menu in the upper right corner and choose Settings.

Copy the URL in the MCP Server section with the copy button next to it.

• Selora Hub: a remote HTTPS URL, such as https://mcp-xxx-xxx.selorabox.com/api/selora_ai/mcp. It works from anywhere, including cloud agents like Claude and ChatGPT.
• Other Home Assistant installations: a local URL, such as http://homeassistant.local/api/selora_ai/mcp. It only works from your home network.

2. Authentication

Selora Connect

The first time your agent connects, your browser opens an Authorize Access screen. Sign in with your Selora Homes account and approve.

Access is read-only by default: the agent can read your devices, states and history, and list your scenes, automations and scripts. Tick Allow control actions to also let it turn devices on and off, change settings and trigger automations.

Selora Hub users have Selora Connect set up already. On other installations, turn on Connect via Selora account in the MCP Server section of the Selora AI settings and follow the steps. You can revoke an agent’s access at any time from the same settings.

Installers can connect their agent to a customer’s hub during a remote support session the homeowner approved. See Connect Your AI Agent (MCP).

Selora AI Token

In the MCP Server section of the Selora AI settings, click Add token under MCP Tokens. Pick the Read-only or Admin preset, or Custom to choose exactly which tools the token may call, and an optional expiration. See MCP Tokens for details.

The token is only shown once: copy it right away.

3. Add the Server to Your Agent Client

Pick your agent and how it signs in.

claude mcp add --transport http selora-ai https://mcp-xxx-xxx.selorabox.com/api/selora_ai/mcp

claude mcp add --transport http selora-ai http://homeassistant.local/api/selora_ai/mcp \
  --header "Authorization: Bearer $SELORA_TOKEN"

{
  "mcpServers": {
    "selora-ai": {
      "url": "https://mcp-xxx-xxx.selorabox.com/api/selora_ai/mcp"
    }
  }
}

{
  "mcpServers": {
    "selora-ai": {
      "url": "http://homeassistant.local/api/selora_ai/mcp",
      "headers": {
        "Authorization": "Bearer ${env:SELORA_TOKEN}"
      }
    }
  }
}

{
  "mcpServers": {
    "selora-ai": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp-xxx-xxx.selorabox.com/api/selora_ai/mcp"
      ]
    }
  }
}

{
  "mcpServers": {
    "selora-ai": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "http://homeassistant.local/api/selora_ai/mcp",
        "--transport",
        "http-only",
        "--header",
        "Authorization: Bearer ${SELORA_TOKEN}"
      ],
      "env": {
        "SELORA_TOKEN": "<your-token>"
      }
    }
  }
}

4. Try It

Ask your agent “analyze my home”. If it describes your home, the connection works.

From there, ask it to propose an automation. Review it before you confirm: new automations are created disabled, and suggestions wait for you to accept or dismiss them.

Troubleshooting

401 Unauthorized with a token: check that the token is complete and the header is exactly Authorization: Bearer <token>. Tokens are only shown once: if you lost it, create a new one.

401 Unauthorized with Selora Connect: turn Connect via Selora account off and on in the Selora AI settings, then connect again.

OAuth redirect error or Access Denied: sign in to the Selora Homes account that owns the installation, in the browser that opens during sign-in.

403 Admin required: your token doesn’t allow this tool. Read-only Selora AI tokens can’t call write tools, and a Home Assistant token must come from an admin account.

Connection errors or timeouts: make sure the agent can reach the URL. Common causes are the wrong port (80 or 8123), a firewall, homeassistant.local used from Docker or another network, or a cloud agent (Claude, ChatGPT) pointed at a local URL.

Wrong automation targeted: ask the agent to list automations first and use the returned automation_id, not a name from memory.

No patterns or suggestions: run selora_trigger_scan, wait for it to finish, then ask for suggestions again.