Using Indigo from a chat client¶
New in 2026.1 Once you've connected a chat client to Indigo (see Connecting a Chat Client), you can talk to your home in plain language. You don't run commands or remember device IDs — you describe what you want, and the assistant uses Indigo's tools to do it. This page explains, in everyday terms, what you can ask for, how Indigo decides which device you mean, what it will and won't do, and the one or two safety steps you'll occasionally run into.
The basic idea¶
Your chat assistant can now see and act on your Indigo home. Behind the scenes Indigo offers the assistant a set of tools; you never call those tools yourself. You just write something like "turn off the porch light" and the assistant figures out which tool to use and which device you meant. If it needs to confirm something with you, it'll ask.
A good way to start is simply:
"What Indigo devices can you see?"
The assistant will list your devices, which is also a quick way to confirm the connection is working and to see how your devices are named.
What you can ask for¶
Check on things. Ask about the current state of your home and the assistant will read it from Indigo:
"Which lights are on right now?"
"What's the upstairs thermostat set to?"
"Is the garage door open?"
"List my variables."
"What action groups do I have?"
Control devices. You can turn things on and off, dim them, set colors, and adjust thermostats:
"Turn on the kitchen lights."
"Toggle the office lamp."
"Set the living room lights to 40%."
"Make the bedroom lamp warm white."
"Set the hallway light to a soft red."
"Set the downstairs thermostat to 68 and switch it to heat."
Indigo controls relays and dimmers (on / off / toggle / brightness), color and color-temperature bulbs (by name, hex, RGB, or white level), and thermostats (heat and cool setpoints, HVAC mode, and fan mode). You can also include simple timing, like "turn the outside lights off in 10 minutes."
Run your automations. Anything you've already built in Indigo can be triggered by name:
"Run my Movie Night action group."
"Execute the Goodnight schedule."
"Fire the Away trigger."
By default, triggers and schedules honor their own conditions — the same as if Indigo ran them itself — so a trigger whose conditions aren't met won't do anything. If you explicitly ask to run something regardless of its conditions, that counts as a sensitive action and the assistant will confirm with you first (see below).
Work with variables. You can read, change, and create variables:
"What's the value of the HouseMode variable?"
"Set HouseMode to vacation."
"Create a variable called GuestArriving set to false."
Ask about your server. There are a handful of system questions too:
"What version of Indigo am I running, and where is it located?"
"What time is sunrise and sunset today?"
"Which plugins are installed, and are they all running?"
"Show me the last 50 lines of the Indigo event log."
"Write a note to the event log that says 'guests arriving at 6'."
Use your tags and folders. If you organize things in Indigo with tags or folders, the assistant can work with those groupings directly:
"What tags do I have?"
"Turn off everything tagged outdoor."
"List the devices in the Upstairs folder."
"Which of my kitchen-area things are on right now?"
Tags cut across object types — the same tag can be on a device, a variable and a schedule — while folders are per-type, so "the Upstairs folder" means a different list for devices than for variables. A tag is just a name; it carries a color that the Indigo clients use to draw it, and the assistant can tell you a tag's color if you ask.
How Indigo knows which device you mean¶
You don't have to name devices exactly. If you say "the kitchen lights," Indigo matches that against your published devices and picks the best fit — and if it's unsure, the assistant can show you the top candidates so you can choose. This works for devices, variables, action groups, schedules, triggers, and control pages.
This matching runs entirely on your own Mac. On an Apple silicon Mac running macOS 26 or later with Apple Intelligence turned on, Indigo uses Apple's on-device intelligence, which is good at casual, indirect phrasing — for example "I'm too warm" can find your thermostat. The macOS 27 model is noticeably better at this than the macOS 26 one. On other Macs Indigo falls back to a built-in word-matching approach, so requests that use the words in an object's name, like "kitchen lights" or "front door lock", usually find it, but indirect phrasing usually won't. The full requirements include a couple that are easy to miss, such as a language setting. Either way, nothing about your home is sent to a cloud service to do the matching.
How often the first match is the right one, in our testing across several real Indigo databases:
| Your Mac | Everyday phrasing | Using names |
|---|---|---|
| macOS before 26, an Intel Mac, or Apple Intelligence off | about 45% | about 80% |
| macOS 26 with Apple Intelligence | about 70% | about 90% |
| macOS 27 with Apple Intelligence | about 80% | about 92% |
"Everyday phrasing" is the way people talk ("the lights in the yard", "is the furnace room flooding"); "using names" is a request that uses most of the object's name ("Office Sonos"). Your own results depend on your names and how you phrase things.
Whichever Mac you have, clear names help most. A name that says what the object is and where it is (Basement Furnace Water Sensor, not Sensor 3) is found far more often than one that doesn't.
If a request is ambiguous, the friendliest fix is to use the name as it appears in Indigo — asking the assistant to "list my devices" shows you exactly how things are named.
What's visible to the assistant (and what isn't)¶
The MCP server deliberately doesn't expose your entire Indigo database. A device, action group, schedule, trigger, control page, or variable is visible to the assistant only if it's marked for remote display — the same visibility setting that controls what shows up for remote Indigo access. Its enclosing folder must be remote-visible too.
The practical effect: if you ask about a device and the assistant can't find it, check that the device (and its folder) is set to be remote-visible in Indigo. This gives you a simple way to decide what your assistant can touch — anything you don't want exposed, just leave out of remote display.
Hide what you never ask about, and finding gets better
Some plugins create a lot of devices: one for every phone and laptop on your network, every Bluetooth beacon, or every sensor reading. They're usually named for the plugin rather than for how you'd talk about them, and every one of them is another name for Indigo to choose among, so they can push the device you meant down the list. Put the ones you'd never ask your assistant about in a folder with remote display turned off. Everything in that folder disappears from the assistant at once, with no need to change each device, and you can move a device back out whenever you want the assistant to see it.
Remote display also controls what remote Indigo clients show, so devices in that folder are hidden there too.
Confirmations for sensitive devices¶
Most actions happen immediately. A few are treated as sensitive and require a second step:
- Locks
- Garage door and door controllers
- Any device you've specifically flagged as requiring confirmation
For these, the action does not fire on a single command. Indigo's MCP server requires a separate, explicit confirmation step before it will act — so a stray, misread, or injected instruction can't throw a lock or open a garage door on its own. In practice this feels like a natural check-in:
You: "Unlock the front door." Assistant: "That's a lock — should I unlock the front door?" You: "Yes." Assistant: "Done."
The two-step requirement is built into the server and is independent of which client you use. The prompt you actually see, though, comes from your chat client's tool-approval UI — that's what puts you in the loop. So if you've told your client to "always allow" Indigo's tools, it will stop asking and carry out the confirmation step automatically. If you want to be asked every time for locks and other sensitive devices, don't grant "always allow" to these tools — in Claude Desktop, review the Indigo connector's tool permissions and clear it. The confirmation token is single-use and expires after a couple of minutes, so a stale approval can't act later.
What the assistant will never do¶
There are firm limits, by design:
- No deleting. The MCP server cannot delete anything in your Indigo database, ever.
- No editing your setup. It can't create or change devices, triggers, schedules, action groups, or control pages — it can only run the ones you've already built. (The one thing it can create is a variable.)
- Only what's published. It acts only on objects marked for remote display, as described above.
Structural changes to your Indigo system stay where they belong: in the Indigo client, done by you.
A few example sessions¶
Evening wind-down
"Dim the living room to 20%, turn off the kitchen, and run my Goodnight action group."
Checking in while out
"Are any doors or windows open? And is the thermostat in away mode?"
Quick fix
"The office is too dark — bump those lights up to 80%."
Housekeeping
"List my variables, then set HouseMode to home."
Talk to it the way you'd talk to a person who can see your Indigo system. When something is sensitive it'll check with you; when something isn't visible it'll tell you; and when a name is ambiguous it'll ask which one you meant.
Troubleshooting¶
These are problems with what the assistant does once it's connected. If it can't reach Indigo at all, or its Indigo tools are missing, see Troubleshooting in Connecting a Chat Client.
It can't find something I know is there. Work through these in order:
- Is it published? The assistant sees only objects marked for remote display, in a folder that is marked too (see What's visible). A hidden object is simply absent, so the assistant can't tell it apart from one that doesn't exist, and may say there is no such thing.
- Is it a device that hasn't finished setup? A device is visible only once it's fully configured in Indigo.
- Does your Mac match by meaning? Without Apple Intelligence, Indigo matches on shared words.
"Turn on the porch light" works, but "I can't see the front steps" finds nothing when no name
contains those words. You can ask the assistant which kind of matching its last search used:
a
tieroflexicalmeans words only. The requirements list what Apple Intelligence needs, including a language setting that hides it without saying so. - Try the name as Indigo shows it. Ask the assistant to list your devices (or variables, action groups and so on) and use a word from the name.
It picked the wrong one. This happens most with names that differ by one word, such as "Porch Light On" and "Porch Light Off" action groups, or a smoke detector on each floor. Use the words that make the difference, or ask the assistant to show you the top matches before it acts. Distinct names help too. The word that only one object has counts for the most: in "turn on the fence lights", "fence" narrows things down far more than "lights", which many of your devices may share. So say that word, and give each object one.
It says nothing matched. The assistant can answer "not found" rather than guess. That is deliberate: running the closest wrong thing is worse than asking. The checks above still apply.
It ran a trigger or schedule, and nothing happened. Triggers and schedules honor their own conditions, so one whose conditions aren't met does nothing. Ask it to run regardless of conditions if that's what you want; it will confirm first.
It acted on a lock without asking. Your chat client has been told to always allow Indigo's tools, so it approves the confirmation step by itself. See Confirmations for sensitive devices.
It won't create, rename or change a device, trigger, schedule or action group. That's by design: the assistant controls what you've built but doesn't change it. See What it deliberately cannot do.