# \[APP\]\[Pro\] 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 **URL:** \ **Category:** Apps\ **Tags:** app, flow, homey-pro, ai, mcp\ **Created:** [June 8, 2026, 4:20pm UTC](https://community.homey.app/t/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/155930 "2026-06-08T16:20:44Z")\ **Posts on this page:** 20\ **Page:** 6
**Author:** ![Mariomark1541](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/mariomark1541/32/178160_2.png) [@Mariomark1541](https://community.homey.app/u/Mariomark1541)\ **Post date:** [September 9, 2026, 6:18am UTC](https://community.homey.app/t/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/155930/102 "2026-09-09T06:18:11Z")
Hi! I just wanted to say that I’m really happy with the MCP AI Bridge app. I set it up recently and it has been working really well for me. The integration with ChatGPT and Homey is incredibly useful. Being able to control devices, check information and even work with Advanced Flows through AI makes Homey much more powerful and flexible. I’ve already been using it quite a lot and I’m impressed by how well everything works. It really opens up a lot of new possibilities for automating and managing my smart home. Thanks for creating and maintaining this app. Great work! 👏 @weide43 ---
**Author:** ![weide43](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/weide43/32/79186_2.png) [@weide43](https://community.homey.app/u/weide43)\ **Post date:** [September 9, 2026, 7:40pm UTC](https://community.homey.app/t/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/155930/103 "2026-09-09T19:40:18Z")
Thanks @Mariomark1541 , that is good to hear. And it is worth saying that this thread got better because of you specifically. The test button that answered “Missing implementation for api endpoint” had been broken since the day that feature shipped, and it would still be broken if you had not written it up with the exact error text. That one is on me, and it is fixed. Good to know ChatGPT and Advanced Flows are holding up. That combination is the one I can test least well from here, so hearing it from somebody actually living with it is worth more than another round of my own testing. If the Ask a model card ever did turn up in the flow editor for you, I would still like to know, so I can stop wondering whether there is something else going on there. There is a fair bit more in the test version now, but that is in the post above rather than here. Enjoy it, and keep the reports coming when something looks off. ---
**Author:** ![weide43](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/weide43/32/79186_2.png) [@weide43](https://community.homey.app/u/weide43)\ **Post date:** [September 11, 2026, 4:41pm UTC](https://community.homey.app/t/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/155930/104 "2026-09-11T16:41:11Z")
## 4.18.0 is in test: the whole app read through, and everything that turned up, built **One thing first: this version needs Homey firmware 12.1.0 or newer.** The app now has a widget for Homey’s own dashboards, and Homey only allows widgets from that version on. On older firmware the update simply will not be offered. ### Custom flow tools actually work now If you built a flow with the card “AI agent calls a custom tool” and your assistant never saw the tool: that was the app, not you. Current firmware stores that card under a longer name than the app was looking for, so no flow ever became a tool. It does now. The tool also follows your flows without a restart: a new flow shows up as a tool within seconds, and the tool of a deleted flow goes away with it instead of claiming the flow ran. A custom tool can now get a description, up to five typed arguments (they arrive in the flow as Argument 1 to 5) and a longer wait for its answer. Ask your assistant to use flow\_tools\_describe. ### Zones and schedules **A room inside a floor counts.** Every zone filter compared one id, so “turn everything off downstairs” did nothing when downstairs was a floor with rooms in it. Sub-zones are included now, and a group change tells you which rooms it covered. **Schedules take a clock time.** “Every weekday at seven” is now a schedule of its own, recomputed against your home’s timezone, so it no longer slides an hour at every daylight-saving change the way an interval does. ### Energy and system Energy reports per day, week, month and year, your energy prices, the state of your Matter and Thread networks, how much each app uses, a backup on request, and installing an app. Installing counts as destructive, so it needs an API key and a confirmation. ### Diagnostics Search the log, filter it by key, and see median and slowest response times, eight days of counts and which tools fail most. How long the logs keep entries is a setting now. There is also a change history that records what something was changed from and to, and by which key. It stays off until you switch it on, because it keeps names and values. ### Flow cards New triggers for a scheduled job that fails and an assistant that disconnects. New conditions for whether an assistant is connected, which security event fired, and whether today’s model spend is below an amount. And two actions that limit the bridge to a lower access mode and restore it. A flow can never raise access above what you chose on the settings page, because an assistant can start flows. The “AI controls a device” card now says which key and which way in did it, and group changes fire it too. The ask card returns what the question cost. ### Dashboards **Every page has its own token.** A dashboard page carries its token inside the page, and until now that token opened every dashboard. So a hallway tablet that only shows lights could, in principle, unlock the front door because another dashboard has the lock on it. Each page now gets a token that opens only that page, and only the controls on it. Pages that were open keep working until they reload. **Live, and honest about it.** The security tab shows live statuses instead of an “all fine” filled in when the page was built. A value from a sensor that stopped reporting is dimmed with its last update, and a page that lost the bridge says since when. **New:** logic variable cards, live charts, position sliders for covers, the outside temperature, flow tiles (behind an API key), and today’s model cost on the ask card and on the Bridge device. And a widget for Homey’s own dashboards with the bridge status. ### Assistants A long action reports progress to clients that can show it, and a request can be cancelled. One device, zone, variable or flow can be read by id instead of reading all of them. One client’s resource subscription no longer sends updates to every other client. Watches go further than a capability: a device going offline, somebody arriving, leaving, falling asleep or waking up, and a logic variable changing. And an assistant building a flow can look up the valid values of a card field that offers a list, such as which HomeyScript to run, instead of guessing. ### What would help Test version: [MCP AI Bridge | Homey](https://homey.app/a/community.mcp-ai-bridge/test/) Everything above was run on a real Homey. A few things I cannot check from here, and I would like to hear about them: - **A custom flow tool you built yourself.** Especially if you had one that never showed up before. - **The widget** on a Homey dashboard: whether it looks right on your screen. - **Model cost** with your own prices filled in, on the ask card and the Bridge device. - **A sensor that goes quiet** on the app’s dashboard: whether the dimming makes sense to you. ---
**Author:** ![Mariomark1541](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/mariomark1541/32/178160_2.png) [@Mariomark1541](https://community.homey.app/u/Mariomark1541)\ **Post date:** [October 2, 2026, 11:26am UTC](https://community.homey.app/t/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/155930/105 "2026-10-02T11:26:05Z")
Still so happy with this app ---
**Author:** ![weide43](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/weide43/32/79186_2.png) [@weide43](https://community.homey.app/u/weide43)\ **Post date:** [October 2, 2026, 11:37am UTC](https://community.homey.app/t/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/155930/106 "2026-10-02T11:37:58Z")
Do you have new suggestions or ideas? ---
**Author:** ![Mariomark1541](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/mariomark1541/32/178160_2.png) [@Mariomark1541](https://community.homey.app/u/Mariomark1541)\ **Post date:** [October 2, 2026, 6:05pm UTC](https://community.homey.app/t/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/155930/107 "2026-10-02T18:05:48Z")
Would it be possible to add something like this for ai flow cards **“Speak AI response”** Ideally, you could choose things like: - which AI voice provider to use - which voice to use - language - speaking speed / style - which speaker should play the result What I mean is a **natural AI-generated voice** , for example using OpenAI TTS or ElevenLabs, rather than the standard Homey/Sonos text-to-speech voice. Example: Dishwasher finishes → AI creates a funny message → the message is spoken on Sonos using a natural AI voice. For example: “Ladies and gentlemen, the dishwasher has successfully completed its mission. Volunteers for unloading may now report for duty.” I think this could also be really useful for doorbell announcements, goodnight messages, weather summaries and other smart home notifications. The nicest implementation would be if the Flow card could take the AI response, generate the audio and play it directly on Sonos/Chromecast, without the user having to create or host an MP3 file manually. ---
**Author:** ![Arie\_J\_Godschalk](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/arie_j_godschalk/32/63027_2.png) [@Arie\_J\_Godschalk](https://community.homey.app/u/Arie_J_Godschalk)\ **Post date:** [October 2, 2026, 8:49pm UTC](https://community.homey.app/t/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/155930/108 "2026-10-02T20:49:25Z")
> [@Mariomark1541](#): > > What I mean is a **natural AI-generated voice** , for example using OpenAI TTS or ElevenLabs, rather than the standard Homey/Sonos text-to-speech voice. Not sure if this is what you want, but the Google Service app lets you select and use many natural voices (TTS) which you can play on your sonos and other devices. ---
**Author:** ![weide43](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/weide43/32/79186_2.png) [@weide43](https://community.homey.app/u/weide43)\ **Post date:** [October 4, 2026, 6:35am UTC](https://community.homey.app/t/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/155930/109 "2026-10-04T06:35:24Z")
@Mariomark1541 Hi Mark, Your “Speak AI response” idea is in the test version, 4.19.0: [MCP AI Bridge | Homey](https://homey.app/a/community.mcp-ai-bridge/test/) 🔊 **Your dishwasher example is one card:** _Ask a model and say the answer_. The flow asks the model for a funny line and the card plays it on your Sonos or Chromecast in a natural voice. Any other answer works too: put the Answer tag of _Ask a model a question_ into _Say a text with an AI voice_. Everything on your list can be chosen per card: - **Provider** : OpenAI, ElevenLabs, Google Cloud, Microsoft Azure, or a voice server in your own house - **Voice** : a picker with that provider’s voices - **Language, speed and style** : “calm, like a butler”, for example - **Speaker** : one speaker, a room, all speakers, or wherever somebody is right now No MP3 to make or host: the app serves the clip on your own network and tells the speaker to play it. A Sonos plays it over the music, and every speaker goes back to its own volume afterwards. Your other examples got cards of their own: _Describe a camera picture out loud_ for the doorbell, and _Say a briefing_ for weather and goodnight messages. To set it up, open the new **AI voice** tab in the app settings, choose a provider, paste your own key and try a test sentence on a speaker. I tested it on a Chromecast with a voice server on my own computer. I have no OpenAI or ElevenLabs key myself, and Sonos I could only test through its card. So I would really like to hear how it sounds with your keys on your Sonos, and if something fails, the exact error text, because that is usually the provider’s own explanation. ---
**Author:** ![Mariomark1541](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/mariomark1541/32/178160_2.png) [@Mariomark1541](https://community.homey.app/u/Mariomark1541)\ **Post date:** [October 4, 2026, 7:36am UTC](https://community.homey.app/t/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/155930/110 "2026-10-04T07:36:55Z")
Wow sounds great! Gonna test it tonight or tomorrow! ---
**Author:** ![weide43](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/weide43/32/79186_2.png) [@weide43](https://community.homey.app/u/weide43)\ **Post date:** [October 4, 2026, 1:22pm UTC](https://community.homey.app/t/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/155930/111 "2026-10-04T13:22:16Z")
Hi Mark, Your idea is in the test version, and I have been using it all day. 🎉 4.19.1: [MCP AI Bridge | Homey](https://homey.app/a/community.mcp-ai-bridge/test/) **I tested it with Microsoft Azure, and it is free.** Azure’s free tier gives 0.5 million characters a month, which is a lot of doorbells and dishwashers, and it never sends a bill: when the month runs out, it simply stops until the next one. I set it up with the Flemish voices Dena and Arnaud, and they sound like a person in the room, not like the computer voice you are used to from Homey or Sonos. From the flow firing to the voice coming out of my Chromecast takes about two and a half seconds, and a sentence it has said before comes from the cache, free and instantly. Your dishwasher example is exactly one card: _Ask a model and say the answer_. The model writes the funny line, the card says it on your speaker, and there is no MP3 to make or host anywhere. Per card you choose the provider, the voice, the language, the speed and style, and the speaker: one speaker, a room, all of them, or wherever somebody is right now. **Setting it up takes a few minutes.** The AI voice tab walks you through Azure in seven steps. Then: - **Test the key** before you save, and the voices appear straight away - choose a voice from the list, with your own language at the top - **Listen** to it on a speaker of your choice right there - **Save and check** tells you in one line what is really stored Prefer OpenAI or ElevenLabs? They work the same way with your own key, and you can even mix them: one card with ElevenLabs for the doorbell, another with Azure for the dishwasher. Want everything to stay in the house? The tab explains a free local voice with Piper or espeak on a PC, NAS or Raspberry Pi. And for your other examples: _Describe a camera picture out loud_ for the doorbell, _Say a briefing_ for the weather and a goodnight, a chime before a message, quiet hours at night, and an intercom on the dashboard. I have tried it on a Sonos myself with Azure, not with OpenAI or ElevenLabs keys. So if you give it a go on your Sonos, I would love to hear how it sounds, and if anything fails, the exact error text helps most. 🙂 ---
**Author:** ![Mariomark1541](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/mariomark1541/32/178160_2.png) [@Mariomark1541](https://community.homey.app/u/Mariomark1541)\ **Post date:** [October 4, 2026, 6:13pm UTC](https://community.homey.app/t/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/155930/112 "2026-10-04T18:13:31Z")
It’s working great also with open AI love this and thank you for quick support ---
**Author:** ![Teddy](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/teddy/32/13818_2.png) [@Teddy](https://community.homey.app/u/Teddy)\ **Post date:** [October 4, 2026, 6:21pm UTC](https://community.homey.app/t/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/155930/113 "2026-10-04T18:21:35Z")
I have been sitting and playing with this over the weekend. Probably way over my head but i decided to give it a try. Ran into a minor problem it seam. This is what i did: I have the test version of the app well i started with the prod version first and i have a PC where i run Ubuntu v26 Server in minimal form with Open WebUI in a docker compose setup. Anyway no problem getting it to work with the AI connections. It say it is all ok toward the MCP AI bridge but i am getting a issue with some type of compability issue between the homey app and WebUI. I am getting a **`WARNING | Unknown SSE event: endpoint`** : not shure what that is but after some searching it seam that it is some type of compability issue with the SSE stream that WebUI doesnt understand and gets interupted and as a result i dont get a result to my question to the AI model. Do you guys have any good sugestion? ---
**Author:** ![weide43](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/weide43/32/79186_2.png) [@weide43](https://community.homey.app/u/weide43)\ **Post date:** [October 4, 2026, 6:24pm UTC](https://community.homey.app/t/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/155930/114 "2026-10-04T18:24:55Z")
Do you have te latest test version? I think you’re busy with the AI Voice feature? But the rest of the app is set up and working? ---
**Author:** ![Teddy](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/teddy/32/13818_2.png) [@Teddy](https://community.homey.app/u/Teddy)\ **Post date:** [October 4, 2026, 6:29pm UTC](https://community.homey.app/t/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/155930/115 "2026-10-04T18:29:57Z")
No im not playing with the voice parts yet. I just wanted to get it to work properly towards homey first. atm i have 4.19.3.test installed but i got the same problem on the prod version. I tried to tell the WebUI to use a older version of the mcp protocol with a yalm environment param but that didnt help eather. Pulled a part of the log out just to show 2026-10-04 17:02:22.181 | INFO | httpx.\_client:\_send\_single\_request:1740 - HTTP Request: POST [http://10.1.40.162:52199/mcp](http://10.1.40.162:52199/mcp) “HTTP/1.1 200 OK” 2026-10-04 17:02:22.181 | INFO | mcp.client.streamable\_http:\_maybe\_extract\_session\_id\_from\_response:181 - Received session ID: 5c30c1d5b20a5605b3a08b4ea320a821 2026-10-04 17:02:22.182 | INFO | mcp.client.streamable\_http:\_maybe\_extract\_protocol\_version\_from\_message:193 - Negotiated protocol version: 2025-11-25 2026-10-04 17:02:22.189 | INFO | httpx.\_client:\_send\_single\_request:1740 - HTTP Request: POST [http://10.1.40.162:52199/mcp](http://10.1.40.162:52199/mcp) “HTTP/1.1 202 Accepted” 2026-10-04 17:02:22.196 | INFO | httpx.\_client:\_send\_single\_request:1740 - HTTP Request: GET [http://10.1.40.162:52199/mcp](http://10.1.40.162:52199/mcp) “HTTP/1.1 200 OK” 2026-10-04 17:02:22.196 | WARNING | mcp.client.streamable\_http:\_handle\_sse\_event:244 - Unknown SSE event: endpoint 2026-10-04 17:02:22.208 | INFO | httpx.\_client:\_send\_single\_request:1740 - HTTP Request: POST [http://10.1.40.162:52199/mcp](http://10.1.40.162:52199/mcp) “HTTP/1.1 200 OK” ---
**Author:** ![weide43](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/weide43/32/79186_2.png) [@weide43](https://community.homey.app/u/weide43)\ **Post date:** [October 4, 2026, 6:45pm UTC](https://community.homey.app/t/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/155930/116 "2026-10-04T18:45:22Z")
@Teddy Hi Teddy, Thanks for the log, it helped a lot. 👍 I ran the same client Open WebUI uses (the official Python MCP SDK) against my own Homey and got exactly your warning. In both your log and mine, everything after it still works: the session is set up, and the last line, the POST with “200 OK”, is the bridge handing over its list of tools. So the bridge is answering. What goes missing is the step after that, between Open WebUI and your model. **The warning.** `Unknown SSE event: endpoint` is harmless. The bridge greets every client the way the older MCP transport expects, and the newer client Open WebUI uses just notes that it does not know that greeting. I have changed it so newer clients no longer get it, and it will be in the next update. While testing I also found that the older transport itself did hang with Python clients, which is probably what you ran into when you told Open WebUI to use an older protocol. That is fixed in the same update, but you do not need it: leave the protocol setting at its default. **The likely reason there is no answer: 173 tools.** Open WebUI sends the description of every tool to the model with your question. 173 descriptions are tens of thousands of tokens, and a local model through Ollama runs with a small context of a few thousand tokens unless you raise it. When it does not fit, the model often just stops without an error. Three things to try, in this order: 1. **Give Open WebUI fewer tools.** In the app settings, on the Security tab under Scoped API keys, create a key for Open WebUI and click Tools next to it. “Read and control” or “Under 100” are quick starts; for a small local model, 20 to 40 tools (devices, zones, flows) work best. Use that key in Open WebUI. 2. **Native function calling.** In Open WebUI, open the model, then Advanced Params, and set Function Calling to Native. 3. **A bigger context.** With Ollama, raise the context length (num\_ctx) of the model, for example to 16384 or 32768, if your PC has the memory for it. If it still stays silent after that, could you tell me which model you use (and whether it runs on Ollama or a cloud API), your Open WebUI version, and the log lines that come after the ones you posted? Then we will find the rest. 🙂 ---
**Author:** ![Teddy](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/teddy/32/13818_2.png) [@Teddy](https://community.homey.app/u/Teddy)\ **Post date:** [October 4, 2026, 7:35pm UTC](https://community.homey.app/t/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/155930/117 "2026-10-04T19:35:45Z")
Lets see i use v0.11.4 on the WebUI. It is Gemini 3.5 to 3.8 flash version i have been playing around with but been consideirng trying something else also. ---
**Author:** ![weide43](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/weide43/32/79186_2.png) [@weide43](https://community.homey.app/u/weide43)\ **Post date:** [October 4, 2026, 7:52pm UTC](https://community.homey.app/t/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/155930/118 "2026-10-04T19:52:07Z")
@Teddy Thanks Teddy, that explains it. 💡 Gemini accepts at most 128 tools in one request, and the bridge offers 173. Open WebUI sends all of them along with your question, Gemini refuses the request, and you get nothing back. Your log stops right after the bridge hands over its tool list, which is exactly where that happens. The fix is already in the app: give Open WebUI a key that only shows part of the tools. 1. In the MCP AI Bridge settings, open the **Security** tab and go to **Scoped API keys**. 2. Create a key, for example “Open WebUI”, and click **Tools** next to it. 3. Under **Quick choice** , pick **Under 100**. That keeps you below Gemini’s limit with room to spare. Fewer is even better: for everyday use, the tools for devices, zones and flows are often enough, and the model picks the right one faster. 4. Use that key in Open WebUI, in your MCP connection as `Authorization: Bearer `. 5. In Open WebUI, set **Function Calling** to **Native** for the model (Chat Controls, Advanced Params). The “Unknown SSE event: endpoint” warning is gone in 4.19.4, which is in the test version now: [MCP AI Bridge | Homey](https://homey.app/a/community.mcp-ai-bridge/test/) It never stopped anything, it just looked alarming. If it stays silent with fewer tools, the Open WebUI log lines right after the tool list usually contain Gemini’s own error message; if you post those, we will find the rest. 🙂 ---
**Author:** ![Teddy](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/teddy/32/13818_2.png) [@Teddy](https://community.homey.app/u/Teddy)\ **Post date:** [October 5, 2026, 4:58pm UTC](https://community.homey.app/t/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/155930/119 "2026-10-05T16:58:19Z")
I think i got a little longer now turned out i had to switch to openAPI and run it over a MCPO proxy. that solved my topic issue and i had to put function calling to legacy. but it stil have something stil beeing off sins i do get a respons not the one i expect. Going to test setting up claude and see how that works tomorrow. If i had it in Native or the standard one i got no tool topics at all just the header structure with nothing in path and the other things. but when putting it in legacy it gave me a crapload 😉 ---
**Author:** ![weide43](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/weide43/32/79186_2.png) [@weide43](https://community.homey.app/u/weide43)\ **Post date:** [October 5, 2026, 6:12pm UTC](https://community.homey.app/t/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/155930/120 "2026-10-05T18:12:25Z")
@Teddy Hi Teddy, thanks for digging into this. What you see in Legacy fits. In that mode Open WebUI pastes all 173 tool descriptions into the prompt and lets Gemini pick one in plain text. That is a lot to choose from, which is why the answers come back a bit off. In Native mode Gemini gets them as real tools, and then it runs into its limit of 128. 4.20.0 went live today and is made for exactly this. You can skip MCPO: 1. Update MCP AI Bridge to 4.20.0. 2. In the app settings, on the Security tab, create a new API key for Open WebUI. Click Tools next to it and tick **Compact mode**. Gemini then sees two tools: tools\_search finds the right action for what you asked, tools\_call runs it. Through those two it still reaches all 173. 3. In Open WebUI: Admin Panel, Settings, External Tools, then +. Type **MCP (Streamable HTTP)**, URL `http://:52199/mcp`, Auth **Bearer** with the new key. Remove the MCPO entry. 4. Set Function Calling back to **Native** in the model’s Advanced Params, and switch the tool on in the chat. The connect wizard in the app now has an Open WebUI tab with these fields filled in for you. If you would rather keep MCPO, give it the compact key: MCPO talks MCP to the bridge, so it will only see the two tools as well. Curious how Claude does tomorrow. Claude takes the full list as it is, so there you can leave compact mode off. ---
**Author:** ![Dominik](https://avatars.discourse-cdn.com/v4/letter/d/ee7513/32.png) [@Dominik](https://community.homey.app/u/Dominik)\ **Post date:** [October 7, 2026, 12:00am UTC](https://community.homey.app/t/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/155930/121 "2026-10-07T00:00:42Z")
Not working ☹ ![image](https://us1.discourse-cdn.com/flex025/uploads/athom/original/3X/c/e/ce1d8f074bc269b50d99e1e80c9bba5282dd6394.jpeg) But App page shows (Homey Pro Early 2019): ![image](https://us1.discourse-cdn.com/flex025/uploads/athom/original/3X/7/8/7892b1fc515f0c4980e8abe17c63f40e5e2db2fe.png) I was hoping that would get rid of the Homey Pro MCP rate limits and to be able to create flows in subfolders. [Previous page](https://community.homey.app/t/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/155930.md?page=5) [Next page](https://community.homey.app/t/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/155930.md?page=7)