Guys,
Am I being a plonker.......... what do i do to even get this started??
i installed this, kept getting killed and out of memory, now i cant get my supervisor / system info tabs to load.
I have been using this for 2 weeks now and it's an awesome tool. doesn't exactly tell you how many tokens you have left in a session but I can check via the actual app.
One thing I did discover is when asking it to help with various config changes, dashboard edits etc, is that it can modify things directly in storage but that doesn't have any ability to update HAs in-memory state. The result is a ton of HA reboots when changing even little things.
I asked it flat out what is needed to allow Claude to do the same edits without the restarts and this is what it said:
a lovelace_get_config and lovelace_save_config (with url_path support for named dashboards) would close the gap completely. the underlying websocket calls are already well-documented in the HA developer docs so it's not a novel feature request - just an exposure gap in the MCP layer.
So that begs the question, is this something that is or is not being considered for this add-on?
does this require any subscriptions?
type “claude” and then hit enter 
So did the buggy makers when the automobile was becoming popular.
So did the whalers when oil was discovered.
So did the film industry when everyone got a digital camera in their phone.
Bookstores - Amazon
Travel agencies - Expedia and Kyak
Encyclopedias - Wikipedia
Taxis - Uber and Lyft
Manual assembly workers - Robots
Typewriters - Word processors
Record stores - Spotify and iTunes
The future is going to be just fine.
Did you ever get an answer?
Are you using the Home Assistant MCP server?
Haven’t gotten an answer, I am connecting my Claude Code instance via the MCP-Server, but I see the advantage of having the claude instance living inside of Home Assistant. You can keep it separate from your other Claude tasks and you can access it easily from your phone or any other Home Assistant capable device. I feel like it comes down to preference and how you like to work.
I am confused (nothing new). Isn’t the MCP-Server giving Claude access to your config? How is that not “living inside of Home Assistant”?
If you want Claude to make an automation, where do you tell it what you want done? In Home Assistant dashboard? In Claude.exe?
Mcp allows access to tools inside HA.
File level access allows all the scripts automations and other things on disk. (this is where it creates an automation if allowed. MCP does not do this)
They are not the same.
MCP is more restrictive - you only get tools that are exposed (and can somewhat be controlled (see v.next of Zenos I’m starting to do exactly this)
File level == Godmode.
MCP allows and MCP does not do this?
Note the confusion. The description of MCP says what it does, but not how to use it.
It depends on what you’re connecting with Steve.
If you’re connecting with Claude (or any MCP client really,) connect the harness to HA using the HA MCP server integration. (Claude opencode, codex et al are clients)
That basically let’s any MCP client ‘see’ the same tools Assist does. The way you do this is up to your harness.
To open file level you use the tool to navigate to your HA config folder and go to town.
When I do my work I use both. A file level connection to write and an mcp connection to test.
Can MCP/Claude be used without Assist?
What’s a harness?
Harness is the program that runs Claude, e.g. Claude.exe, opencode or codex. (It has a facility to connect to mcp, each is slightly different but basically the same connection.
Yes. What it connects to are your tools. (exposed scripts and intents) and not exactly the list of tools exposed to assist is the de facto list of tools exposed to MCP.