ChatGPT Homey integration - “Too many requests” rate limit

Questions & Help
1

Hi everyone,

I’m using a Homey Pro (2026) connected to ChatGPT through the official Homey integration.

I regularly use ChatGPT to inspect, create and modify my Homey Flows and Advanced Flows. Unfortunately, I keep running into the error “Too many requests”, which makes the integration quite frustrating to use.

Today, for example:

  • ChatGPT attempted to access a single Advanced Flow and immediately received the error.
  • After waiting, we tried again with just one request, but received the same error.
  • I also refreshed the Homey connection in ChatGPT, without success.
  • No changes could be made to my Flows.

What makes this particularly frustrating is that the error occurs even when making just one request, without sending multiple requests in quick succession.

I understand that API rate limits are necessary, but the current limitation seems quite restrictive for users who want to actively manage their smart home through ChatGPT.

My questions:

  1. Is there a known rate limit for the official ChatGPT–Homey integration?
  2. How many requests are allowed, and over what period?
  3. Is this limitation imposed by Homey or ChatGPT?
  4. Is there a way to check when the limit resets?
  5. Are there any plans to increase the limit or make it more flexible?

I really like the possibilities this integration offers, especially for managing more complex Advanced Flows. However, the recurring rate limit makes it difficult to use reliably.

I’d love to hear whether others experience the same issue and whether there’s a solution or improvement planned.

Thanks in advance!

2

My advice : use the mcp offered as a community app. It’s not as limited as the official one, it works locally and you don’t have obviously rate limits

3

Do not use the official MCP Server! It is a useless toy and has obviously been abandoned by Athom a long time ago.
There are very good custom apps for LLM integration around. Check these out. They usually work pretty well.

4

@Bkinsbergen I can’t tell you what the limit of the official integration is. That one runs through Athom, so only they can answer those questions.

What I can tell you about is the alternative the others mean. I made MCP AI Bridge, so I’m biased: it runs an MCP server on the Homey itself, and ChatGPT talks to that directly. There is no rate limit unless you set one yourself on the Security tab. It works on the Homey Pro (2026).

For flows, which is what you use it for:

  • ChatGPT can read, create and change basic flows and Advanced Flows. Before every change the app keeps a copy of the flow, and one call puts the old version back.
  • It can show what an edit would change before it happens, check a flow before it is saved, and propose a repair for a flow that broke after a device was removed.
  • You decide per connection what ChatGPT may do: only read, operate devices, or also change flows.

What you need: ChatGPT Plus or Pro with developer mode on, because ChatGPT only takes custom connectors there. ChatGPT runs on the internet, so your Homey needs a public HTTPS address. The Remote tab in the app sets that up in four steps: a free DuckDNS name or your own domain in Cloudflare, a Let’s Encrypt certificate the app fetches and renews itself, one port forward in your router (443 to 52200), and a check. Then you add a custom connector in ChatGPT with that address.

The app topic, with the full setup: [APP][Pro] MCP AI Bridge: talk to your Homey with Claude, ChatGPT, Gemini, Copilot, Cursor, Mistral, Ollama, LM Studio, n8n, Home Assistant or any AI, also from Telegram and WhatsApp

5

And where did you get that information from? That’s simply not correct. We added new features to our MCP integration just last week, and development is very much ongoing.

I understand that you might be frustrated, and that’s perfectly fine. However, frustration shouldn’t be a reason to spread incorrect information, especially when people like @Bkinsbergen are genuinely looking for answers or trying to understand the current state of things.

I think it’s important to understand a little about how MCP actually works. MCP (or Model Context Protocol) is an open standard that allows AI assistants like ChatGPT and Claude to interact with external services, including Homey.

Essentially, our MCP server provides the AI assistant with a collection of tools that it can use to interact with Homey. These tools allow it to retrieve information about your devices, control them, start Flows, and even create or modify regular and Advanced Flows.

However, Homey doesn’t control how the AI assistant interprets your requests or which tools it decides to use. We provide the tools and describe what they do, but it’s ultimately the AI model that decides how to use them. This also means that different AI models, or even different conversations with the same model, can produce different results.

That’s also where a lot of our ongoing development comes in. We’re continuously looking at how we can make these tools easier for AI models to understand by providing the right context and make more of Homey’s functionality accessible through MCP (Just like Variables, which was added last week).

6

You forgot to explain the “Too many requests” error that they get.

7

Sorry for me being wrong. I must have overlooked that information. Where do I find it?
Just had a look at ✨ Homey MCP Server | Megathread but didn´t see any news there.

8

That could be caused by a number of things. One thing I noticed myself, for example, was that when requesting Insights data, ChatGPT didn’t know what resolution to use. That’s something we can improve by providing more specific context to help ChatGPT understand what information it needs to request.

With Flows, it could be something similar. For example, ChatGPT might be retrieving a list of all Flows instead of requesting a specific one, or searching through folders without being able to find the Flow it’s looking for. These are all things we can investigate and potentially improve.

@Bkinsbergen; could you share the exact request you made to ChatGPT, or perhaps the relevant part of the conversation? That would help us understand what’s happening and where we can improve the integration. Via a DM is also fine of course.

9

Fair point, but I don’t think that’s a reason to claim we’ve abandoned development :wink:

A changelog would be a nice addition though. I’ll pass that feedback along.

10

A changelog would be handy. It’s easy to miss improvements when there’s no obvious place to follow them.

On the original issue, is there a way to tell which service is returning “Too many requests”? Sharing the prompt should help, but knowing which request failed and whether ChatGPT made several calls along the way would make it easier to narrow down.

11

Today and yesterday I have been adding a lot of devices to homey, and if you add devices, than you want to use them in a smart way and thus need flows. I have been making 10 to 15 flows today using chatgpt, and for a simple flow it all works well. But when it becomes just a little bit more complex (and I’m not talking anything crazy here), I keep running in this “Too many requests” issue. chatGPT even learned himself that it can try a couple of more times after waiting a bit. It is really disruptive, and I am really trying nothing crazy here. Just some everyday flows with a couple of devices.

Scherm­afbeelding 2026-11-02 om 19.52.44
Scherm­afbeelding 2026-11-02 om 19.52.44748×250 17.9 KB

Scherm­afbeelding 2026-11-02 om 19.57.39

12

Hi @Doekse,

Thanks for your detailed explanation and for looking into this! Apologies for the slightly delayed response.

I can share a concrete example from October 28.

I was using ChatGPT to optimize my existing downstairs lighting automation. We were working on an Advanced Flow called LIGHT 2026 - WidgetKeeper centrale lichtmodus.

During this process:

  • The first attempt resulted in a “Too many requests” error.

  • After waiting, ChatGPT tried again and specifically attempted to retrieve the Advanced Flow mentioned above.

  • That request also returned “Too many requests”.

  • Refreshing the Homey connection didn’t resolve the issue.

  • No modifications could be completed.

I’ve noticed that this happens particularly frequently when working on more complex automations involving multiple Advanced Flows.

Sometimes ChatGPT can successfully inspect and modify several Flows before suddenly encountering the limit. Other times, the error occurs almost immediately, even when starting a new task.

One important thing I’d like to emphasize is that I already try to prevent unnecessary API requests by giving ChatGPT very specific instructions and limitations.

For example, I regularly instruct ChatGPT to:

  • Only inspect or modify specific, explicitly named Advanced Flows.

  • Avoid performing full audits or unnecessary inventories.

  • Reuse existing Flows, scripts and variables instead of creating new ones.

  • Avoid repeating previously completed work.

  • Focus exclusively on the requested changes and minimize unnecessary operations.

Despite these instructions, I still regularly encounter the “Too many requests” error, sometimes before ChatGPT can even retrieve the specific Flow I’m asking it to work with.

Of course, I understand that providing these instructions doesn’t necessarily prevent ChatGPT from making multiple API calls internally. However, it makes me wonder whether the integration could provide better guidance to the AI model to help it work more efficiently.

My Homey setup currently contains approximately 159 Advanced Flows, including some fairly complex automations involving multiple devices, variables, timers and scripts. While this is a reasonably large setup, I wouldn’t expect its overall size to cause rate-limit issues when ChatGPT is instructed to work on just one specific Flow.

I also understand from your explanation that ChatGPT decides which tools to call, and that a single user prompt doesn’t necessarily mean a single API request.

This raises a few questions:

  1. What are the actual rate limits? Is there a fixed number of requests allowed per minute or hour?

  2. What triggers the limit? Is it based on the number of API calls, the amount of data retrieved, or something else?

  3. Could ChatGPT be making unnecessary requests despite explicit instructions? For example, retrieving entire Flow lists when only one specific Advanced Flow is needed.

  4. Is there a way to see when the limit resets? Ideally, ChatGPT would receive a clear retry time instead of repeatedly encountering the same error.

  5. Could the integration be optimized for larger Homey setups? Especially when working with multiple existing Advanced Flows.

What @Tom_Tieman describes is very recognizable. The integration works really well when it works, but these interruptions make it difficult to complete more complex automation tasks reliably.

I would particularly appreciate more transparency about the limits and how to avoid triggering them. Right now, it’s difficult to know whether I should simply wait, start a new conversation, or approach the task differently.

I’m happy to share the relevant ChatGPT conversation or additional examples via DM if that would help your team investigate. It would also be useful to know whether the error originates from Homey’s MCP server or from ChatGPT itself.

And thanks for confirming that development of the official integration is ongoing! I genuinely enjoy using it and would prefer to continue with the official Homey integration rather than switching to a third-party alternative.

Thanks again for your time and efforts!