Connect GlacierGrid to Claude

Last updated: 29 July 2026 · Privacy policy · Support

GlacierGrid monitors and manages HVAC, refrigeration and energy systems across your sites. This connector lets you ask Claude about that data in plain language - why a freezer is alarming, which sites are overspending, whether your temperature records hold up for an audit - and get an answer built from your live equipment data instead of a dashboard you have to interpret yourself.

GlacierGrid manages your equipment; this connector only observes it. As it stands today the connector reads your GlacierGrid data and nothing else - it cannot change a setpoint, acknowledge an alert, create a work order, or message your team, even though the platform itself can do those things. If that ever changes, anything that modifies data will be marked as such and Claude will ask before running it.

Before you start

You'll need:

• An active GlacierGrid account. If you sign in to GlacierGrid today, you're ready.
• At least one site with reporting hardware - sensors, thermostats or gateways. The connector reads existing telemetry; without devices reporting, most questions return nothing.
• Claude on web, desktop or mobile. Custom connectors work on every Claude plan including Free, though Anthropic limits Free accounts to one - so if you already have a connector added, you'll need to swap it or upgrade. Plan terms are Anthropic's and can change; their connector guidance is the current word on it.

One thing worth knowing up front: GlacierGrid scopes people to the locations their team covers. If a colleague sees different sites than you do, that's your GlacierGrid permissions working as configured, not a fault in the connector.

Connect it

1. In Claude, open Settings → Connectors → Add custom connector.
2. Paste the connector URL:

   https://mcp.glaciergrid.com/mcp

3. Claude opens a GlacierGrid sign-in page. Choose Sign in with Google or use the email and password form - whichever you normally use for GlacierGrid. The page also shows an Internal (GlacierGrid Staff) button. That one is for GlacierGrid employees only and will reject a customer account.
4. Approve the connection. Claude returns to your conversation with GlacierGrid connected.

You won't be asked to create an API key or paste a client secret - the connection is authorised against your existing GlacierGrid login and refreshes itself.

To disconnect, remove the connector in Claude's settings. Claude stops using it immediately.

Try these three prompts

These exercise different parts of the connector. Paste one in after connecting.

1. Investigate a site problem

One of our walk-in freezers keeps alarming. What's going on?

Claude finds the unit, checks whether its sensor is actually reporting, reads the recent temperature history, pulls the open and recently resolved alerts, and looks at how the unit has been trending - a sudden excursion reads very differently from a slow drift. Naming a site - "the walk-in freezer at our Riverside store" - skips straight to it if you have several.

What you'll get back: the most likely cause with the evidence behind it - the temperature trace against the unit's configured thresholds, how many times it has alarmed recently and for how long, whether the sensor has gone offline (a gap in data, not a warm freezer), and whether the pattern looks like a one-off excursion or something recurring - plus a recommended next step.

Walk-in Freezer — Riverside · currently 10.4–13.1°F against a 10°F threshold
Active alarm opened 11:30 AM local (food safety) · 58 alarms in the last 30 days
Dips to 7–9°F in normal operation, then spikes to 12–20°F every 2–4 hours and self-recovers
Most likely: defrost cycle timing, a door seal, or condenser coil condition
Bottom line: a persistent control problem rather than nuisance alarming — worth a service call

(Claude then offered to break the spikes down by time of day to separate a defrost pattern from door traffic.)

2. Find where you're overspending

Which of our sites are overspending this quarter, and what's driving it?

Claude ranks your sites by energy savings performance, then looks at what's behind the worst performers: how hard zones are working to hold their setpoints, runtime and equipment behaviour, and the outdoor conditions each site faced.

What you'll get back: sites ranked by savings percentage, with each underperformer paired to a likely driver - a zone struggling to hold temperature, excessive runtime, or equipment not delivering the cooling it should. If nothing is actually overspending, it says so rather than manufacturing a problem to match the question.

No site is in net overspend — every location with billing data is saving against its predicted baseline, so the real question is which are capturing the least.
Weakest five: 1.4–2.1% savings, against a fleet median nearer 8–10% — roughly 6–8% left on the table at sites spending six figures a year on energy.
Directly linked: four sites have active HVAC faults, including one running stage-2 cooling at a 4.5°F supply-air delta — maximum power draw for minimal cooling.

(The current quarter had no billing data, so Claude widened the window to 24 months itself and then offered a comfort-burden analysis on the weakest sites.)

3. Produce a temperature-monitoring record

Give me a temperature-monitoring record for our refrigerators and freezers for last month.

Claude assembles a per-unit record against your configured food-safety thresholds.

What you'll get back: for each unit, how much of the period was actually monitored, how much of that time was in range, every sustained out-of-range event with its start time, duration and peak, and any stretch where monitoring went dark long enough to matter.

30-day record across 326 monitored units — 160 freezers, 164 refrigerators — each measured against the threshold configured for it.
Per unit: how much of the window was actually monitored, what share of that time was in range, and every sustained excursion with its start, duration and peak.
Flagged: units that never held their threshold all month, and multi-hour excursions well above it — separated from the routine 60-minute defrost spikes, which are expected.
Coverage gaps: 12 sensors reported for under 60% of the window — stretches where compliance is not "in range" or "out of range" but simply unknown.

(Claude then offered to map the worst units to specific store locations for dispatch.)

This is a record of what your monitoring captured - useful to keep on file and to hand to an auditor. It measures air temperature at the unit, not the temperature of the food itself, and food-code compliance is assessed on product temperature by an inspector. So treat it as evidence for your records, not a food-safety determination, and not a replacement for your own food-safety programme (HACCP or equivalent).

Because the connector doesn't write to GlacierGrid, what it produces is the report itself. There's nothing to save back into the platform, and nothing it changes while answering.

Guided reports

Alongside free-form questions, the connector ships prepared workflows. In Claude they appear as prompts you can pick rather than type.

Store Complaint Investigator answers tersely first and offers to go deeper - that short first reply is intentional, not a truncated response.

What the connector can see

Scoped to your company, and read-only as it stands today:

• Your sites, equipment and installed devices
• Temperature and humidity readings from your sensors
• Thermostat state, setpoints and runtime behaviour
• Open and historical issues and alerts
• Who is configured to receive alerts, and which notifications were sent
• Energy and cost savings by site and billing period
• Outdoor weather for your sites' locations, for comparison

Your team's contact details are hidden by default and only included if you ask for them directly.

It has no access to anything outside GlacierGrid, and doesn't modify anything inside it.

Known limitations

• Temperatures are shown in °F and times in each site's local timezone. Ask if you'd prefer Celsius.
• A device that stops reporting shows as a gap, not a zero. Devices are treated as offline after they've been silent for about an hour, and Claude will say so rather than present a stale reading as current.
• Refrigeration trend analysis needs at least two weeks of history. Below that, slow degradation can't be told apart from normal variation.
• Energy savings data lags utility billing by weeks. Recent periods may be empty.
• Cycle and runtime analysis is available for supported smart thermostats. Sites on other equipment still get sensor, issue and alert data.
• Account-wide questions across many sites can take up to a minute. Claude is doing the aggregation server-side rather than pulling every device individually.
• Requests are rate-limited. Normal use won't hit it; rapid repeated querying may pause briefly.

Troubleshooting

Privacy, security and support

The connection is authorised against your GlacierGrid login using OAuth - GlacierGrid never sees your Claude account, and Claude never receives your GlacierGrid password. Access is limited to your own company's data, and every capability currently offered only reads.

Disconnecting the connector in Claude's settings stops access immediately.

Read our privacy policy for what we collect, how long we keep it, and how to request deletion. For help, contact support@glaciergrid.com.