ChatGPT reaches AppFolio or Yardi data only through a connector, such as a custom connector to an MCP server that reads them. With user-level access, users sign in with their own AppFolio or Yardi account (or your SSO), and answers inherit their existing property and portfolio permissions. Agree occupancy, delinquency and rent definitions with accounting first, and expect every answer to show its query.
The question behind the search
Property managers ask the same questions every week. What is occupancy by property? Who is delinquent past 30 days? Which leases expire in the next 60 days without a renewal offer? Which vendors take longest to close a work order? Those answers live in AppFolio or Yardi, and getting them usually means running a report, exporting it and reading the spreadsheet.
So people ask whether ChatGPT can just look. It can, if something connects them. ChatGPT and Claude reach outside systems through connectors, and a custom connector can point at a remote MCP server. The server calls the property management system and returns the result. Everything worth checking sits in that server and whose access it uses.
How AppFolio and Yardi are reached
Both systems expose data to integrations through interfaces the vendor controls, and both already decide which properties and portfolios each user can see.
| System | Interface | What decides each user’s access |
|---|---|---|
| AppFolio | The Reports API, plus partner APIs in the AppFolio Stack program | The user’s AppFolio role and property access |
| Yardi | Interface web services, such as the Voyager SOAP interfaces licensed through the Yardi Standard Interface Partnership Program | The user’s Yardi role and property list |
Access follows the user
With user-level access, a regional manager who sees only her portfolio in AppFolio or Yardi sees only her portfolio in an answer too. Nexatron inherits those existing permissions, so the property system stays the authority on who sees which properties.
Yardi interfaces are licensed. Confirm which interfaces your database is licensed for before planning questions, because a question about data the interface does not expose has no answer. With AppFolio, check which reports the Reports API returns for your plan, since the question list should follow what the API can actually read.
Definitions accounting will ask about
Property reporting is full of terms with two defensible meanings. Settle these with accounting and property management before comparing an assistant’s answer with an owner report.
- OccupancyPhysical, by occupied units, or economic, by rent collected against potential rent. Decide whether units on notice count as occupied.
- DelinquencyBalance by aging bucket on the tenant ledger, whether prepayments offset it, and which charge types count.
- Scheduled rentLease rent on occupied units, or market rent on every unit.
- Vacancy daysFrom move-out to the next move-in, or to the next lease start.
- PortfolioWhich properties belong to each portfolio or owner, as of which date.
Who may ask, and about which properties
With user-level access, each user’s existing property permissions decide what an answer covers, and the connecting platform decides who may ask at all. Questions can also be limited to approved definitions. Test it with two users who should see different portfolios, and confirm each gets only their own.
Owners, tenants and vendors are a separate matter. Their portals have their own access, and nothing in an internal assistant should widen it. Start with staff whose access in AppFolio or Yardi is already well understood, and widen the group once the test passes.
Setting it up in ChatGPT
On ChatGPT plans that allow custom connectors, a workspace admin usually controls whether they can be added. A person adds the MCP server’s URL, signs in to it, and enables the connector in a conversation. Claude follows a similar pattern with remote MCP servers. OpenAI and Anthropic both document the current steps, and menu names change often enough that it is worth reading them at setup time.
Start with one narrow question, such as physical occupancy for one property on one date, and compare it with the same report in AppFolio or Yardi. Then try a question that should be refused, such as a property outside the user’s portfolio, and a request to change a record, which should never succeed.
Questions worth starting with
A short list makes the first week useful and the comparison honest. Start with physical occupancy for each property on a named date, delinquent balance over 30 days as a share of monthly scheduled rent, leases expiring in the next 60 days with no renewal offer, and median days from work order creation to completion by vendor last quarter.
Each of these has a matching report in AppFolio or Yardi, so accounting can check the answer against something it already trusts. Keep the list for later reviews, because a change in the source system or the connection can quietly change an answer that matched before.
How Nexatron answers property questions
Nexatron connects to AppFolio through the Reports API and to Yardi through its licensed interfaces with user-level access: users sign in with their own account (or your SSO), and Nexatron inherits their existing property permissions. People ask in the Nexatron app or from ChatGPT, Claude and Microsoft Copilot through Nexatron’s hosted MCP server. Questions only read the property system.
Each answer names the reading it used for words such as occupied or delinquent, with the other readings one click away, and shows its query. It reads Verified only once a named person on your team has reviewed the logic. A question Nexatron cannot answer under your rules is declined with the reason and what to do next. The AppFolio rent roll and work order pages list example questions and the definitions to agree.
Claude is a trademark of Anthropic, PBC.
Bring three business questions
Explore your sources, definitions, and review criteria with our team. Please keep credentials and sensitive records out of the inquiry.