LLM ChatGPT Agent MCP API Integrations

Hi ​@daveps63 

Welcome to the Sonos Community!

Thank you - I've marked this thread as a feature request and it will be seen by the relevant teams for consideration. Keep the ideas coming!

OP, I’m curious to hear what your use case is. I’m not doubting there is one, just want to know what you’re thinking.

Since Sonos already has an API, and AI can help you create a custom app/script to access your Sonos system.  It sounds more like you want to you want to be able to say/speak a command to your AI of choice and have it carry out the commands live? I  can see some benefit to that, over just using an app or voice assistant.  Was there something else you were thinking of doing?

I have been creating IOS Shortcuts which may not be as versatile as AI.  Just wondering how you might Group all rooms then set the volume for each room then choose the content to play.  How would that be done using LLM?

I’m using Claude (far from an expert) but you can setup reusable skills that are very flexible in terms of how you use them.  They are also easy to setup.   In the context  of Sonos for example, you could setup a command “/sonos_party  XXXXX” (where XXXXX is the content) that would know to group all rooms, set a volume that is appropriate for the time of day and content selected.  You can let Claude decide what is appropriate, possibly with additional context or .  This isn’t something you can do that easily do without several steps of the app or voice command.   I don’t even think this is something Alexa + can do since it doesn’t seem to have any memory or hold a lot of context of your specific environment, but I honestly haven’t used it that much.

But I’m not OP and don’t know if that’s what was in mind.  If the goal is just to create a controller app, or  write automation through some other third party, I don’t know that an MCP is needed. 

​@melvimbe 

In Shortcuts I have a shortcut that selects coaxial on the KEF LS50 WII speakers (Port connected ), sets the KEF volume, groups all 5 Sonos Rooms, sets a volume for each Room, chooses content, turns on the Shuffle variable & turns on two mood lights.  The request is “Hey Siri. Richard”.

A Pause shortcut ungroups, sets volumes back to standard settings & turns OFF mood lights.

if it’s a boring group of folks at the party I have a shortcut named Low that drops all the volumes….Yawn.

I would love to try Claude someday.  A friend of mine uses Claude to great affect.

A full MCP server would make Sonos much more flexible. Being able to control rooms, playback, volume, and groups through LLM agents would be a great upgrade.

 You can use Claude for free, but it does have a limit of how much you can use it before you run out of your processing allotment and have to wait a set of time to use it again.  I’m using a free account at home, and have successfully got all the issues with my HA setup all fixed.  Now I have to new features to address.   I have considering using Claude to build my own app, but I can’t really come with a feature I need that would be worth the effort.  I have used it for some basic productivity.

I do use Claude Code at work and that’s been very productive.  One thing about using these LLMs as there isn’t really a wrong way to use them.   Yes, there are better strategies for organizing, spending less tokens, and getting results faster, but it will work with whatever strategy makes sense to you will you ramp up. 

Any specific example use cases you have in mind?   The way I see it, you can create automations through several different means without an MCP, and even use LLMs to create those automations.  Where LLMs would add value is when you want an LLM to make a decision/assumption about what to play without your direct input.  I’m generally curious to know.