What you can ask.

• Before you start
• Name your station
• The words that work
• Your sensors
• Right now
• The forecast for your spot
• Rain
• Watering
• Yesterday and any past day
• History
• Hour by hour
• Records, streaks and counts
• Watches
• More than one station
• Questions that combine
• What it can't answer yet
• If something goes wrong
• Test checklist
• For developers

Before you start

You need an Ambient Weather station that reports to AmbientWeather.net and is connected to Station Signals; a plan with agent access (Silver, Gold or Platinum; the Free plan doesn't include it, and paid plans aren't open yet, so join the wait list to hear first); and an AI agent that connects to remote MCP servers with a sign-in. We've tested it with Claude and Claude Duet, OpenAI Dots, Gemini, Grok Bot and Meta Muse. The steps below are for Claude. The others each have their own place to add a custom connector, and the address is the same.

To connect Claude

1. In Station Signals, open Settings and find “Your AI agent.” Copy the address: https://stationsignals.app/mcp.
2. In Claude, open Settings, then Connectors, and add a custom connector with that address.
3. Claude sends you to Station Signals. Sign in, pick your household and tap Allow. The screen lists exactly what the agent can read before you say yes.
4. Back in Claude, start a new chat and make sure Station Signals is switched on for it. Then ask: "What's the temperature in my yard?"

To disconnect, open Settings in Station Signals and tap Disconnect next to the agent. It stops working on its next question. The same list shows when each agent was connected and when it last asked. If you don't use an agent for 60 days, its connection lapses and you connect it again.

Name your station

Every station has a name, and the name is for your agent. It starts as Yard. In Station setup, pick the word you'd say out loud: Yard, Garden, Vineyard, Orchard, Farm, Ranch, Pasture, Greenhouse, Cabin, Lake house, Beach house, Barn, or something else you type. When you ask “Did it rain in the garden?”, that name is how the agent knows which station you mean.

• One word, the word you already use. If you call it the back forty, name it “Back forty.” The agent repeats the name in its answers, so pick one you'd like to hear.
• Skip people's names and street addresses. Ambient's own label for the device (often a family name) is kept beside the station as “Ambient device,” for telling devices apart. It isn't the name your agent answers to.
• Names are unique within a household. Two stations can't both be Garden.

“The garden,” “my garden” and “garden” all work. So do “my station” and “my weather station.”

The words that work

Say “my yard” or your station's name. “My station,” “my weather station,” “my AWN” and “my Ambient” work too. These words tell the agent to read your station instead of searching the web or pulling a city forecast. “What's the temperature?” on its own may get you a city forecast. “What's the temperature in my yard?” gets you your station.

Your sensors

Every plan with agent access can ask these.

• “What sensors does my station have?”
• “Are all my sensors working?”
• “When did my pool sensor last report?”
• “Do any of my sensors have a low battery?”

What you get back: every sensor your station sent readings from in the last 24 hours: the outdoor array and the indoor console, extra temperature and humidity channels 1 to 10, soil, leaf wetness and leak sensors, the lightning detector, outdoor and indoor PM2.5, the AQIN indoor air monitor and CO2, and relays. For each you get what it reports, whether it's reporting now, when it was last heard from, and its battery. Channels carry the names and roles you gave them, like Pool or Shade.

Good to know. The Ambient API has no sensor list, so Station Signals works out the list from which readings arrive; a pool float shows as “Channel 4: temperature,” plus your name for it. A sensor silent for 30 minutes shows as not reporting, with the time it was last heard from; one silent for more than 24 hours drops off the list. Batteries show as OK or low where the sensor sends a reading Station Signals knows how to read.

Right now

These come from your station's latest reading, usually less than five minutes old. Every plan with agent access can ask them.

• “What's the temperature in my yard?”
• “What does it feel like outside?”
• “How humid is it in my yard? What's the dew point?”
• “How windy is it? Which way is the wind coming from?”
• “What's the strongest gust in my yard today?”
• “Is the pool warming up?”
• “Is it raining in my yard right now? How hard?”
• “How much rain has my yard had today? How much this storm?”
• “When did it last rain in my yard?”
• “What's the barometric pressure? Is it rising or falling?”
• “What's the UV index in my yard right now?”
• “How strong is the sun right now?”
• “When's sunrise and sunset?”
• “What's the temperature inside the house?”
• “What's the pool temperature?”
• “How warm is it in the shade?”
• “When was the last lightning, and how far away was it?”
• “What's the air quality in my yard?”
• “Are there any weather alerts for my yard?”

What you get back: the reading and the time it was taken. Readings count as stale when the station hasn't reported for a while, and the answer says so. Pressure comes as sea-level pressure (what the TV weather uses) and the raw pressure at your station, plus the change over the last 3 hours. Pool and shade come from the extra sensors you named; lightning needs a lightning detector; air quality needs a PM2.5 sensor, and the answer gives the US EPA Air Quality Index with its category. Without the sensor, the answer leaves it out instead of guessing.

The forecast for your spot

This is the National Weather Service forecast for your station's exact location, not the nearest city. Your station measures; it doesn't forecast. The forecast is NWS's, and the answer says so.

• “Will it freeze in my yard tonight?”
• “How cold will my yard get tonight?”
• “Will it rain tomorrow?”
• “What's the forecast for my yard this weekend?”
• “What will the temperature be in my yard at 6 PM?”
• “Is it going to be windy on Saturday?”
• “What's the 7-day forecast for my yard?”

What you get back: an hour-by-hour forecast (temperature, chance of rain, wind, humidity and dew point) for up to 156 hours, a day-by-day forecast for 7 days, and any active alerts. The agent asks for 48 hours unless you want more.

The part no weather app does: how your yard runs against the forecast. Station Signals saves the NWS forecast issued before each day begins, then compares it with what your station measured. Ask “Does my yard usually run warmer or colder than the forecast?” or “Adjusted for my yard, how cold will it get tonight?” Once at least 3 days are compared, the answer includes the average gap for highs and for lows, and every forecast day comes with your yard's high and low already adjusted. If the adjusted low is 36°F or under, the answer calls it near freezing, and 32°F or under a freeze, even when the NWS number alone looks safe. Up to 30 days are used; a new station needs 3 full days before the averages appear.

Rain

Rain comes from your own gauge.

• “How much rain did my yard get in the last 24 hours?”
• “How much rain this month?”
• “How much rain this rain season?”
• “How much rain in the last 12 months?”
• “How many days since it rained in my yard?”
• “How much rain did my yard get each month this year?”

Good to know. Your rain season starts when you say: January 1 (calendar year) or October 1 (water year), in Settings. If your station joined partway through a window, the total says when counting started, so a 2-week-old station never passes off 2 weeks of rain as a year's. There is no lifetime total.

Watering

Gold and Platinum. The same advice as the Watering Coach on your display, for the yard, lawn, garden, trees and desert plants.

• “Should I water tonight?”
• “Can I skip the sprinklers this week?”
• “When do the trees need water again?”

What you get back: for each planting, water tonight, hold off because rain is likely, or skip until a given day, with how far short it is and how much it uses a day. It works out the water each planting used from your station's own temperature, humidity and wind (the FAO-56 method irrigation districts use), takes off the rain your gauge caught, and checks the NWS chance of rain ahead. It needs 3 full days of your station's history, and it assumes you water when it says to: it doesn't know when your sprinklers last ran.

Yesterday and any past day

A plain-sentence recap of one day, with the numbers behind it, for any day within your plan's history reach.

• “How was yesterday in my yard?”
• “What happened in my yard on Saturday?”
• “What was the high and low on October 20?”
• “When did it get coldest last night?”
• “Did it rain yesterday? When?”
• “Did yesterday beat the forecast?”
• “Did yesterday set any records?”

What you get back: the high and low with their times, the most it felt like, how much rain fell and in what spells, the top gust with its time, what the NWS forecast said before the day began, any record the day set that still stands, dry days in a row, and how much of the day the station recorded, flagged when it's partial. A real one from our station:

High 79° at 2:45 PM. Low 52° at 7:00 AM. No rain. 3 dry days in a row. Top gust 9 mph at 12:55 PM. The forecast said 77°. Your yard hit 79°. Coldest night since this station started.

Your week, month, season and year

The same plain-sentence recap for a whole week (Monday to Sunday), month, season or year. Week and month on every paid plan; season and year on Gold and Platinum, with the same window last year beside it. Platinum agents also hear in their regular check when a new recap is ready.

• “How was last week in my yard?”
• “How wet was October at my station?”
• “How did this summer compare with last summer in my yard?”
• “Give me my yard's year.”

History over days, months and years

• “What were the highs and lows in my yard last week?”
• “How much rain did my yard get in October?”
• “What was the average high in my yard last month?”
• “Compare this October with last October in my yard.”
• “Show me my yard's year, month by month.”

By day: high, low, average, rain, peak gust, average humidity and peak UV; partly recorded days are flagged, not hidden. By month or year: days recorded, the highest and lowest temperature, average high and low, total rain, rain days and peak gust. Up to 400 days in one day-by-day answer; longer than that and the agent groups by month or year. History reaches back to the day your station joined Station Signals.

Hour by hour

Gold and Platinum. The last 30 days, including today so far, up to 7 days per question.

• “When did the rain start on Tuesday?”
• “What time did it hit 100 today?”
• “How fast did it cool off after sunset last night?”
• “What was the temperature at 3 AM?”

Records, streaks and counts

Records come in three kinds: this year, all time (since your station joined), and by calendar month. The records are the hottest day, the coldest night, the wettest day and the strongest gust; a record that was beaten shows the one it replaced.

• “What's the hottest day my yard has had this year?”
• “What's the coldest night on record in my yard?”
• “Is this the coldest November night my yard has had?”
• “How many dry days in a row has my yard had? What's the longest dry spell?”
• “How many 100-degree days has my yard had this year?”
• “How many freezing nights this year?”

Streaks: dry days, rain days and 100°F days in a row, each with its current run and its best. Counts per year: days 100°F or hotter, 110°F or hotter, freezing nights, days with rain, days with half an inch or more, and days with a 30 mph gust. Silver sees this year's records and counts; Gold and Platinum see everything.

Watches: have your agent check your yard

Gold and Platinum. A quick check built for an agent that runs on a schedule. It reports what is past a line right now (cold, heat, a big gust, rain falling, lightning within 10 miles in the last 30 minutes, an NWS alert, a low battery, or a station quiet for 15 minutes or more), what is coming in the next 24 hours by the NWS forecast, and what happened since the last check: rain starts and stops with amounts, the high and low, the peak gust, lightning, and gaps in the readings.

Default lines: under 35°F, over 110°F, a gust over 40 mph, and any rain. Set your own in the request:

• “Check my yard and tell me if anything needs attention.”
• “Watch my yard tonight and warn me if it drops below 32.”
• “Tell me if the gusts in my yard go over 25 mph.”
• “Check my yard every morning at 6 and tell me what happened overnight.”

If your agent can run a task on a schedule, ask it to check your yard every hour or every morning, and to tell you only when something needs you. Each answer includes a “next since” time; passing it back on the next check means nothing is missed between checks. A check can look back as far as 7 days. The forecast gives sustained wind, not gusts, so a forecast wind flag means steady wind over your line.

More than one station

Every plan includes 1 station. On Silver, Gold and Platinum, each extra station is an add-on. Give each station its own name, mark one as home, and name the station in your question: “What's the temperature at the cabin?” “Compare the highs at the yard and at the farm yesterday.” Without a name, the agent uses your home station; “my station” and “my yard” mean the home station too.

Questions that combine

• “Should I cover the plants tonight?” Tonight's low for your spot, adjusted for how your yard runs.
• “Is it a good day to swim?” Pool temperature, air temperature, UV and the afternoon forecast.
• “Do I need to water this week?” On Gold and up, the watering advice above. On Silver, rain in the last 7 days and the chance of rain ahead.
• “Write me a one-paragraph weather summary of my yard this week.”

What it can't answer yet

• History for add-on sensors. Shade, indoor, air quality and lightning are right-now only. The pool has one exception: the current reading says whether it is warming, cooling or steady, with its change over the last day and week.
• Pressure, wind direction, dew point and solar history. History covers temperature, rain, gusts, humidity and UV.
• Soil moisture, leaf wetness and leak sensors. Not read yet.
• Snow. Rain gauges don't measure snow depth.
• Anything before your station joined Station Signals.
• Changes. The agent can read, never change. Names change in Station setup.
• Units. Answers come in °F, inches, mph and inHg. Ask the agent to convert, and it will.

What it never sees: your Ambient API key, your station's exact location, other households, and anything it could change. Names you typed come back marked as typed text; the agent treats them as names, never as instructions.

If something goes wrong

• “Agent access comes with a paid Station Signals plan.” Your household's plan doesn't include agents right now.
• “get_attention comes with Gold plans and up.” Watches need Gold. Ask “What's happening in my yard right now?” instead.
• “get_watering comes with Gold plans and up.” Ask “How much rain has my yard had this week?” instead.
• “The station's location isn't set yet.” Set your location in Station setup.
• “This household hasn't connected a station yet.” Connect your station in Station setup first.
• “No station called ... This household's stations: ...” Use a name from the list, or leave the name out for the home station.
• “The station couldn't be read just now.” Try again in a minute. Ambient limits how often a station can be read.
• The agent searched the web instead. Say “my yard” or your station's name, or start a new chat with Station Signals switched on.

Test checklist

Ten questions that cover every tool. Run them in order in a new chat.

For developers

A 403 on every call, even right after signing in. Station Signals answers a bad sign-in with 401, never 403. The hosting in front of the site refuses the default Python user agent before the request reaches us, so a script built on urllib needs to send its own User-Agent, like MyAgent/1.0.

Questions about any of this go through the contact page.