When configuring a conversation agent, you have to select from either controlling Home Assistant or just one of your many MCP integrations. It seems to be a bug - is this just a UI limitation (i.e. should be a checkbox where you can select multiple items, than a radio button)? I’m expecting conversation agent to be able to connect to multiple MCP servers and also retain ability to control Home Assistant.
Still a work in progress, but I have it setup so my conversation agent is a n8n workflow. With the new MCP nodes in n8n, I can essentially create or use as many MCP servers as I want. I currently have Home Assistant, Telegram, Gmail, SerpAPI, and OpenWeatherMap, and chat memory running in my workflow. Still has some kinks to work out but it functions pretty well so far. I’m working on multi-task commands and LLM tweaks ATM.
Hi Nathan,
Thanks for your reply! Just to clarify, when you say “call the other MCP agents from the primary agent that has Assist”, what exactly do you mean by “call”? Are you referring to tool calls through the LLM or some internal routing?
I’ve tried interacting with Assist, but it doesn’t seem to have visibility into the MCP modules I’ve installed. Am I missing a step in the configuration?
You will use a script that invokes a second conversation using the agent or tool of choice. You can track the conversation I’d and reuse it if you need to continue the conversation.
Basically agent A needs internet agent b has internet. A calls b and says hey internet for me and send me your answer.
This is an interesting workaround, but it does add a lot of complexity to achieve what should be a relatively basic functionality. While it’s great that you’ve found a way to make it work with the current tools available, having conversation agents call other agents introduces unnecessary complexity to an already complex system.
The ideal solution would be an official Home Assistant-supported method that simplifies this process and allows users to easily connect their conversation agent to multiple MCP servers. This would provide a more streamlined and user-friendly experience, without requiring extensive custom scripting and workflow management.
