MCP enabled n8n workflows should have description

[HubertReX]

As I understand, n8n workflows that have MCP access enabled should have a description. See below:

This is a hint for an AI tool that wants to use it on how it works and what parameters should be provided. I'm not sure if it's a bug, since CAAL, to some extent, is still able to use them. Still, I believe it would work better with a description. I haven't found workflow descriptions in the repo. Tried to write them on my own, but still, it would be better to get the original intent from the author 😉

[cmac86]

@HubertReX, maybe there is conflicting documentation, and if there is please let me know. But CAAL actually receives its tool instruction from the Notes within the Webhook node. For some reason the description is not passed through MCP. Originally was going to use description but ran into the wall, so some of the workflows may have a description that looks like instructions. I could clean that up.

So all the included workflows in the n8n-workflows folder should have descriptions in the webhook notes.

Home Assistant Control Tool Example:

n8n_create_caal_tool Example:

[HubertReX]

oh, ok, I see it. I have an idea to use Claude Code with n8n a bit differently. Let me explore it, and I'll get back to you.

[NotAlreadyG]

Caal is not correctly using the tool, I mean It's showing in the UI that the tool it used is correct(of n8n) but still it shows some error occured or errors like request.tools something

[cmac86]

@NotAlreadyG, can you include the errors here?

docker logs caal-agent —tail 100

[NotAlreadyG]

Turn detection: interruptions=True, endpointing_delay=0.5s

Wake word: ENABLED (model=models/hey_jarvis.onnx, threshold=0.5)
STT instance type: WakeWordGatedSTT
STT capabilities: streaming=True
Session STT: WakeWordGatedSTT
Loading OpenWakeWord model from models/hey_jarvis.onnx
OpenWakeWord model loaded
{"level": "WARNING", "name": "livekit.agents", "message": "no request_id provided for TTS livekit.plugins.openai.tts.TTS", "pid": 241, "job_id": "AJ_QLoiU7j9dWsv", "room_id": "RM_x9U7uq5v85C2", "timestamp": "2026-01-20T09:04:52.361062+00:00"}
Agent done, follow-up window started
Agent ready - listening for speech...
LLM returned 1 tool call(s)
Executing tool: light1_on with args: None
Calling n8n workflow: light1_on
Failed to execute n8n workflow light1-on: 500, message='Internal Server Error', url='https://username.hf.space/webhook/light1-on'
Error executing tool light1_on: 500, message='Internal Server Error', url='https://username.hf.space/webhook/light1-on'
Traceback (most recent call last):
File "/app/src/caal/llm/llm_node.py", line 424, in _execute_tool_calls
tool_result = await _execute_single_tool(agent, tool_name, arguments)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/app/src/caal/llm/llm_node.py", line 495, in _execute_single_tool
result = await execute_n8n_workflow(
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/app/src/caal/integrations/n8n.py", line 132, in execute_n8n_workflow
response.raise_for_status()
File "/app/.venv/lib/python3.11/site-packages/aiohttp/client_reqrep.py", line 636, in raise_for_status
raise ClientResponseError(
aiohttp.client_exceptions.ClientResponseError: 500, message='Internal Server Error', url='https://username.hf.space/webhook/light1-on'
Error in llm_node: tool call validation failed: attempted to call tool 'hass_control{"action": "turn_on", "target": "main light"}' which was not in request.tools
Traceback (most recent call last):
File "/app/src/caal/llm/llm_node.py", line 146, in llm_node
async for chunk in provider.chat_stream(messages=messages, tools=tools):
File "/app/src/caal/llm/providers/groq_provider.py", line 162, in chat_stream
async for chunk in stream:
File "/app/.venv/lib/python3.11/site-packages/groq/_streaming.py", line 146, in aiter
async for item in self._iterator:
File "/app/.venv/lib/python3.11/site-packages/groq/_streaming.py", line 192, in stream
raise APIError(
groq.APIError: tool call validation failed: attempted to call tool 'hass_control{"action": "turn_on", "target": "main light"}' which was not in request.tools
{"level": "WARNING", "name": "livekit.agents", "message": "no request_id provided for TTS livekit.plugins.openai.tts.TTS", "pid": 241, "job_id": "AJ_QLoiU7j9dWsv", "room_id": "RM_x9U7uq5v85C2", "timestamp": "2026-01-20T09:05:02.903120+00:00"}
Agent done, follow-up window started

[NotAlreadyG]

[cmac86]

A few things going on here:

1. light1_on returns 500 - Your n8n workflow is failing. Check the execution logs in your HF Space or the n8n execution to see what's broken. If the workflow is failing, see what node it fails at and what the error message is.

2. hass_control not found - The model is trying to call hass_control but it's "not in request.tools" Are you connected to Home Assistant in the CAAL settings? The default prompt mentions hass_control, but if you're using custom tools like light1_on instead, you should update your prompt to remove references to hass_control to avoid confusion.

The model sees the prompt say "use hass_control", tries your custom tool first, it fails, then tries to fall back to what the prompt says - but that tool doesn't exist.

What model are you using?

3. The name confusion screenshot- I realize I haven't injected the name from settings into the prompt. I need to fix that. And probably tweak the prompt so it's not so focused on creating tools.

[NotAlreadyG]

Everything else solved, but that wasn't just name problem after using it for a few minutes and making it to call only tools, it forgot that he has a LLM model I mean idk, it was saying don't have a tool for that want me to create one? everytime I asked it anything like even time or it's name or even hi?