Alarmanlagenmanager

Connecting ChatGPT and Claude

What works afterwards

A connected assistant can – depending on the granted scope – for example:

"What's the status of my alarm system?" "Turn on the hallway light." "Create a rule: hallway light on for 60 seconds when the front door opens." "Are there any faults?"

The assistant reads and switches exclusively through the Alarmanlagenmanager – with exactly the permissions you gave it at sign-in.

Connecting via connector (MCP)

The Alarmanlagenmanager is an MCP server (Model Context Protocol – the open standard through which assistants use tools). The address:

https://alarmanlagenmanager.de/api/v1/mcp

In ChatGPT: Settings → Apps/Plugins (depending on the interface also "Connectors") → add custom connector → enter the address. On some account tiers developer mode must be enabled once under the advanced settings; it can be switched off again after adding.

In Claude (claude.ai or desktop): Settings → Connectors → add connector → enter the address.

Signing in

On the first command the assistant shows a Sign in button:

  1. You land on the Alarmanlagenmanager sign-in page and sign in with your account (including the code if 2FA is active).
  2. You choose the panel and tick the scope (switching, automations, arm/disarm, detectors, settings – reading is always included).
  3. Link – done. The assistant is ready.

The consent page only offers panels and scopes your account itself is allowed to use – a restricted user cannot give their assistant more than they have themselves.

Arming/disarming in chat

For switching the alarm mode the Alarmanlagenmanager requires the system's user code with every command – the assistant asks for it in the chat and only passes it through. It is never stored. So arming stays protected even if someone else got hold of the chat history or the assistant account.

Managing and disconnecting

All links appear in the panel view under AI access → Connected services – with account, service, scope and last use. Disconnect ends the access (fully at the latest when the current one-hour token expires). The assistant itself also offers removing the connector; disconnecting here works independently of that.

Alternative: GPT action with a key

If you run your own GPT with actions in ChatGPT (today only possible in business/enterprise environments), connect the system via the OpenAPI schema https://alarmanlagenmanager.de/api/v1/openapi.json and an AI key as bearer authentication. For your own programs and scripts the key is the right route anyway.

Limits

  • The assistant doesn't invent device names: it works with the real device list. If it mistypes a name, it receives the valid names in response and corrects itself.
  • Before saving a rule it shows the plain-language preview – only your confirmation saves.
  • Whatever the scope doesn't allow is rejected – with a message pointing to the permission management in the Alarmanlagenmanager.