±icon->Homey. You do need the “Homey” ChatGPT app of course, because you need to link it to your account
The only App I have installed is the Homey Docs Companion which I have used to setup Claude to audit and document my smart home environment.
I am using ChatGpt simply via the browser using the link that I was taken to by following prompts in the email from Athom. It totally knows my environment and has written disabled enhanced flows to document a strategy it developed to rewrite one of my very complex flows.
I have connected ChatGPT to Homey.
This is really fantastic. However, whenever ChatGPT creates a flow, it keeps showing that Homey gives the message: “Too many requests.”
Can this be solved?
Anyone got a good place or md file with all apps, devices, When, and, then cards, tags, how thigns connects, common gotchas and so on?
I’m using Claude to build pretty complex automations in microsoft environments in different languages, building MCP servers and what not but Claude can barely create a flow worth the name.
So I’m guessing it lacks in training and the kb I’ve created so far isn’t much.
I’ve had it read all the official pages but yeah, it’s still pretty stupid. 
I would like the AI to start flows with a parameter (mainly a string in most of my cases). The current API/tool „start_flow“ does not support this. - Could you enhance the tools to start flows with a parameter?
ChatGPT complains the the JSON it gets for an advanced flow, with about 100 cards, is getting truncated.
The ability to create and fine-tune Advanced Flows using ChatGPT through MCP is really impressive.
For example, I can now create custom OpenAI prompts that analyse tvo independent solar PV systems, evaluate its current state, estimate the expected production and potential energy exports at specific times, and then automatically deliver the result to the relevant people in a clear, human-readable form.
All of this combines data from several fantastic Homey apps developed by the community, together with Advanced Flows, HomeyScript and AI-based interpretation.
Simply: wow. I honestly did not expect this level of flexibility and integration. Very well done!
I tested the Homey MCP server some time ago and posted feedback about a few limitations and issues I encountered.
Is there any way to know whether the MCP server has been updated since then? For example, is there a changelog, version number, release notes, or a list of recently fixed issues somewhere?
Or is the only reliable way to find out to reconnect the integration and repeat the same tests?
I would also be interested to hear whether anyone has recently retested functionality that previously did not work properly, and whether you have noticed any improvements.
Turns out that grok does not receive a truncated flow, so it seems like it’s an issue with ChatGPT.
Do you use the webapp or desktop app with Work or Codex? For stuff like that use only the last ones or work in web/app. The context limit in the chat is smaller, especially for people who don’t pay.
Coming back to the brief discussion in 🎙️ The Homey Podcast | Megathread with @Doekse about Athom / Homey & AI:
Instead of opening a new thread as recommended I´d like to add to this one and ask Athom to at least give their MCP sever some additional love.
Some enhancements would be extremely helpuful for a deeper Homey / AI integration:
Useful stuff for automations:
- trigger_flow_with_argument
- get_energy_*
- list_moods, set_mood
- list_alarms, set_alarm
Useful stuff for AI supported diagnostics:
- list_apps
- get_zwave/matter/zigbee_log
Maybe some scary yolo stuff as well:
- reboot_homey, restart_app
When I recently suggested to support, that a tool to start flows with an argument (trigger_flow_with_argument) would be great, I got the advice to create a web hook, have the AI call this and do whatever in a Homey flow. - That is of course valid and a “killer argument” with which half of Homey could be made unnecessary.
I also know, that workarounds for many of the above can be implemented.
But nevetheless a more complete MCP implementation would be helpful for those of us that are trying to do something useful with AI right now.
