Anthropic's Claude can connect to Hummingbird property management software in about 40 minutes, with no software development background. The first endpoint to hit is the management summary report. Within a week, operators running this setup have scaled the same connection across full portfolios.
This guide walks through that setup as a repeatable sequence. It is built from workflows independent operators are using in production, plus the current documentation for the two products involved: Claude Cowork (Anthropic's agentic workspace that can read local files and drive a browser) and the Nectar open developer API from Tenant Inc., the layer that exposes Hummingbird data to third-party tools.
The point is not that AI is clever. It is that the setup is short, cheap, and boring once you know the order of operations. Here is the order.
What Do You Need Before You Start?
Three things. A Claude plan that includes Cowork (Anthropic ships it on Pro, Max, and Enterprise). A Hummingbird PMS account with Nectar developer access from Tenant Inc. And a spare computer you can wipe.
That last item is not optional, and it is the step most people skip. Run the agent on an old laptop you wipe clean and dedicate to this purpose. Keep it plugged in, always on, always connected. Reach it from your phone through Dispatch, Claude's phone-to-desktop feature, and keep a remote-access tool like TeamViewer installed as a fallback for the rare terminal task.
Why a dedicated machine? Because Claude Cowork can control the computer it runs on. The security logic is straightforward: use a computer you can wipe, and only put into it the logins, access, and data you want the agent to have. An isolated machine turns a powerful, slightly risky tool into a contained one. For a deeper look at why that isolation matters, see our technical teardown of AI operations in self-storage.
How Do You Get API Access to Hummingbird?
Create a Nectar developer account and request keys for a single facility first. Go to Tenant Inc.'s Nectar developer portal, then ask your team for API keys scoped to your facilities. Start with one store to test the connection before scaling.
Nectar is Tenant Inc.'s open API. It is the same connector that lets vendors like StoragePilot, Lumio, and Patchwork Labs read and write live Hummingbird data, documented in our coverage of the StoragePilot Hummingbird integration. For an operator connecting their own AI instead of a vendor product, Nectar is what makes the model's requests hit real occupancy and revenue numbers rather than a static export.
Scope the keys narrowly at first. One facility, read access. You want to confirm the plumbing before you hand the agent your full portfolio or any write permission.
What Are the First 40 Minutes, Step by Step?
The core connection is the fast part. Break it into six moves:
- Open Claude Cowork on the isolated machine and give it the project folder or workspace you will use for storage operations.
- Point Claude at the Nectar API documentation. Log into the developer screen so Claude can read the API docs and connect to Hummingbird. Claude reads the docs and tells you what it needs.
- Hand it the API keys when it asks. Claude will request keys, you supply them, and the connection proceeds from there.
- Hit the management summary report endpoint first. This single report exposes occupancy, move-ins, move-outs, revenue versus budget, and web and phone leads. It is the highest-value read in the system.
- Handle the occasional terminal step. The extent of "coding" here is sometimes copying a command into a terminal window and pasting the reply back. Claude tells you exactly what to paste.
- Confirm the connection returns live data. At this point the model can see every data point on the management summary report for that one store.
That is the 40 minutes. Everything after this is refinement, not connection.
How Do You Make Sure It Is Not Hallucinating?
Verify before you trust, then build memory. This is a required step, not a nicety.
Pull the same report from Hummingbird by hand and check the model's numbers against it, line by line, for one period. Only after the numbers reconcile should you expand scope.
Then give the model history and memory. Have Claude read 12 months of management summary reports and build a knowledge base so it remembers the baseline between sessions instead of re-deriving it every time. Once one store is verified and remembered, adding the rest is just more keys: supply the keys for each additional facility and confirm the data loads correctly.
What Should You Build First, and What Comes Later?
Build the monthly report you actually want, then automate its delivery. The reporting layer needs read access only, so it is the safest high-value starting point. Iterate with Claude until the report matches how you like to read it: separated by brand, rolled up to the portfolio, with custom fields like return on ad spend. Then schedule it: every month Claude can deliver a PDF of that exact report via iMessage or email.
From there, three additions extend the same foundation:
- A market feed. Connect TractIQ through its official MCP connector so your monthly report includes rent-per-square-foot and supply pressure for your target markets. That connector is the same one described in our piece on the TractIQ and CRED iQ AI data connector.
- Segmented reports. Build a different KPI report for your VP of operations and district managers, delivered to a separate channel. Same data, different slice.
- A real-time listener. For faster back-and-forth, replace a polling "watchdog" with a "listener" that responds to text messages in near real time.
Save write-back and customer-facing automation for last, and gate them behind explicit permissions.
The Setup Checklist Worth Saving
- Dedicated, wiped machine running Claude Cowork, always on, reachable via Dispatch
- Nectar developer account with API keys scoped to one facility to start
- Claude pointed at the Nectar API docs; keys supplied on request
- First endpoint: management summary report
- Numbers verified by hand against the live report before scaling
- 12 months of history loaded and a knowledge base built for memory
- Read-only by default; write-back enabled deliberately, per task
- Scope expanded to the full portfolio only after single-store verification
Short Setup, Long Payoff
The connection is a 40-minute job. The discipline around it, isolation, verification, and permission scoping, is what separates a reliable analyst-on-call from an expensive liability writing to your live ledger.
Start read-only. Start with one store. Start with the management summary report. Verify the numbers with your own eyes, build the memory, and only then widen access and turn on write-back. Operators who follow that order get the payoff without inheriting the risks that come from skipping guardrails. The tooling is finally simple enough that the setup is no longer the hard part; the judgment is.
Sources
- Claude Cowork Overview, Anthropic
- Nectar Data Connector, Tenant Inc.
- TractIQ Launches AI Connector, Bringing Verified Self-Storage Data Into Claude and ChatGPT, TractIQ