Nabu Casa–hosted AI / MCP endpoint for secure AI integrations

[aero-oli]

Describe the feature

I’d like to suggest a first-party, Nabu Casa–hosted AI / MCP (Model Context Protocol) endpoint that enables secure integration between Home Assistant and external AI / agent platforms (e.g. ChatGPT Apps SDK or similar).

The core idea is that instead of every user exposing their own Home Assistant MCP endpoint (with a unique *.ui.nabu.casa URL and long-lived tokens), Nabu Casa could provide a single, stable cloud endpoint (for example https://mcp.nabucasa.com/mcp) that:

• Authenticates users via their Nabu Casa account
• Routes requests internally to the correct Home Assistant instance
• Enforces permissions and scopes centrally
• Prevents users from having to share URLs or access tokens with third parties

This would allow AI platforms to integrate with Home Assistant through a single fixed connector, while still respecting per-user identity, permissions, and privacy.

It effectively makes Nabu Casa the secure bridge between AI agents and the user’s Home Assistant, instead of pushing that responsibility onto third-party developers or end users.

Use cases

This would specifically enable Nabu Casa to build and operate an official “Home Assistant for ChatGPT” app using the OpenAI Apps SDK, backed by a single fixed MCP endpoint.

Because AI app platforms require a single stable public endpoint per app, a Nabu Casa–hosted MCP service would make it technically possible to offer a first-party ChatGPT integration that:

• Lets users connect their Home Assistant to ChatGPT by simply logging into their Nabu Casa account
• Avoids asking users to paste URLs or long-lived tokens into third-party services
• Allows Nabu Casa to control authentication, permissions, rate limiting, and auditing centrally
• Ensures that all AI access to the home is opt-in, scoped, and revocable

This would unlock experiences such as:

• Asking ChatGPT questions about the home (“Which windows are open?”, “Did any motion happen last night?”)
• Debugging automations conversationally (“Why didn’t the heating turn off yesterday?”)
• Safely running AI-assisted actions with user confirmation (“Create an automation to turn off lights when everyone leaves”)
• Exploring and managing Home Assistant in natural language, without exposing internal APIs

In short, this would allow Nabu Casa to offer a secure, privacy-respecting, first-party ChatGPT app — instead of leaving this space to unofficial community proxies and integrations.

Anything else?

This feels like a strong fit specifically for Home Assistant Cloud because:

• Nabu Casa already operates the remote access proxy and identity infrastructure
• It already manages subscriptions, trust, and security expectations
• It can define a consistent permission and privacy model instead of fragmenting the ecosystem

Without this, the ecosystem will likely fragment into multiple third-party proxy services with varying security standards, and users will be pushed toward unsafe practices like pasting long-lived tokens into external services.

This could be included as part of the existing subscription or as a higher-tier “AI / agent integration” feature, and positioned as the official, secure way to connect Home Assistant to modern AI tools.

I’m happy to help test, design, or iterate on this idea if it’s useful.