Add changing URL for OpenAI service

[profucius]

Great addon, it works great in my setup. I have two different AI services I want to connect, Google Gemini which I confirmed works, and an alternative OpenAI compatible service, which has its own custom URL.

Other services can offer their LLMs through an "OpenAI compatible" protocol, which is to say that it's the same as OpenAI but you can change the URL and route calls through it instead of actual OpenAI. Think of it kinda like how Ollama is handled, but it is using the OpenAI protocol instead.

I am asking if you would please add this feature to MCP Assist. I imagine it either as a new service in the dropdown selector box, or as a toggle checkbox after selecting OpenAI in the dropdown box. When indicated, the user can add their own URL, and everything else should be the same.

Thanks again for a great addon and I'm excited to see how it grows from here!

Search - What is OpenAI Compatible

Example docs - Open WebUI on configuring OpenAI Compatible service

[mike-nott]

Cool suggestion, I've made the URL field editable rather than preset, so you can your own custom one.

[profucius]

@mike-nott

Thanks for this! I updated and tested it out, but I'm having an issue reaching the next step.

I added a service, selected "OpenAI", added my URL, added a fake API (my URL doesn't require one), but the Model Name box doesn't pull down the model selections from the URL. I manually pasted the exact model name (from another application that I use for this same purpose, but not home assistant related, so I know that it does work).

Then I started a chat, and got an error immediately:

The model 'gemini-3-flash-preview' wasn't found on OpenAI. Check that the model name is correct in your integration settings.

(ignore the name of that model, it is running locally and confirmed correct in another application)

Is this because MCP Assist isn't connecting to the service through the new URL?

[mike-nott]

@profucius this should hopefully be fixed now, please can you test?

[profucius]

Just tried 1.15.3 but the same issue occurred.

^^^ Error about the API field from previous step, which isn't required for my workflow (any token or null works)

^^^ Also the model field does not populate with choices if I delete the text

I think MCP Assist may not be reading the OpenAI Compatible API? Would it help if I gave an example or provided more documentation on how the OpenAI Compatible API works?

[mike-nott]

Would it be possible to get some debug logs showing:
🌐 FETCH: Starting OpenAI model fetch
📡 FETCH: Requesting OpenAI models
📥 FETCH: OpenAI response status XXX

[profucius]

Ok update, in the process of gathering debug logs for you, I found in the log it mention Testing basic streaming to: http://192.168.1.135:2048/v1/v1/chat/completions which tells me it was assuming the /v1/ and I was including it in my URL. I've taken that out now, and I've made some progress towards solution.

Now when I chat with the service, it doesn't throw the error, and it does send the request to the API, however the request is sent just as this text: User: turn off office heater but nothing happens. The chat window stalls out waiting on a reply from AI, however the AI did finish already.

Chat in Home Assistant, stalled out waiting

API request completed already

I'm currently focused on identifying the specific "office heater" entity ID. My approach involves using the "Andrew Office" area index to narrow down the search. I'm planning to utilize the discover_entities command with the area parameter set to "Andrew Office" to find the target.

---

Update again, the chat in Home Assistant finally failed after being stalled out. I gathered the logs from that, posted below:

2026-02-01 16:02:49.937 ERROR (MainThread) [custom_components.mcp_assist.agent] ❌ Streaming iteration 1 failed: 
2026-02-01 16:02:49.937 WARNING (MainThread) [custom_components.mcp_assist.agent] Streaming failed (), using HTTP fallback
2026-02-01 16:03:20.937 ERROR (MainThread) [custom_components.mcp_assist.agent] Error processing conversation
Traceback (most recent call last):
  File "/usr/local/lib/python3.13/site-packages/aiohttp/client_reqrep.py", line 539, in start
    message, payload = await protocol.read()  # type: ignore[union-attr]
                       ^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.13/site-packages/aiohttp/streams.py", line 703, in read
    await self._waiter
asyncio.exceptions.CancelledError

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/config/custom_components/mcp_assist/agent.py", line 644, in _async_process_with_chatlog
    response_text = await self._call_llm(messages)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/config/custom_components/mcp_assist/agent.py", line 2001, in _call_llm
    return await self._call_llm_http(messages)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/config/custom_components/mcp_assist/agent.py", line 2069, in _call_llm_http
    async with session.post(
               ~~~~~~~~~~~~^
        url, headers=headers, json=clean_payload
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    ) as response:
    ^
  File "/usr/local/lib/python3.13/site-packages/aiohttp/client.py", line 1510, in __aenter__
    self._resp: _RetType = await self._coro
                           ^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.13/site-packages/aiohttp/client.py", line 779, in _request
    resp = await handler(req)
           ^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.13/site-packages/aiohttp/client.py", line 757, in _connect_and_send_request
    await resp.start(conn)
  File "/usr/local/lib/python3.13/site-packages/aiohttp/client_reqrep.py", line 534, in start
    with self._timer:
         ^^^^^^^^^^^
  File "/usr/local/lib/python3.13/site-packages/aiohttp/helpers.py", line 713, in __exit__
    raise asyncio.TimeoutError from exc_val
TimeoutError

[kaikaikai86]

successfully configured a locally hosted openAI target. My openai interface runs on lightLLM. This works here perfectly. Great! thanks!

[horizon99]

I could also successfully connect MCP Assist to Swiss Infomaniak's AI service. The base URL is https://api.infomaniak.com/2/ai/{my_product_id}/openai and I manually set the model to qwen3
Thanks for this feature!