ChatGPT + Cadastre

OpenAI's assistant, through a custom MCP server added as a plugin on the web. Connect it to Cadastre, the data source for real-estate data except listings, and it can screen and rank all 393 U.S. metro areas, read metro profiles and compare them, with every figure's source attached.

Works with Cadastre through: the MCP server.

Quick start

1. Add Cadastre as a custom MCP server

On the web, go to ChatGPT Plugins (opens in a new tab), select the plus button, then Add custom MCP server. Enter the name Cadastre and this Server URL, with no authentication, review the warning, select I understand and want to continue, then Create as a plugin, and install it.

https://getcadastre.com/mcp

That is all: no account or sign-in is needed for Cadastre's public datasets, at 30 requests a minute. For higher limits, see Sign in for higher limits.

2. Ask

In a chat, type @, choose the Cadastre plugin, and ask, for example:

Use Cadastre to find metros with population over 500K, manufacturing share of GDP over 8% and unemployment under 5.5%. Rank them on manufacturing share and cite the sources.

Sign in for higher limits

Optional. To sign in, connect through the server's sign-in address in place of https://getcadastre.com/mcp (MCP clients sign in only when the server asks them to):

https://getcadastre.com/mcp?auth=required

Create the plugin with the sign-in address as its Server URL and choose OAuth for authentication (with dynamic client registration). ChatGPT opens Cadastre's sign-in when you create or connect the plugin.

Your browser opens Cadastre's Connect an app page. Sign in to Cadastre (or join first), choose the account ChatGPT should work in, and select Approve. ChatGPT keeps its own access and renews it: there is no API key to copy.

The higher limits need API access turned on for the account: the page marks any account without it, and it cannot be chosen. Write to support@getcadastre.com to ask for it.

MCP configuration

If you sign in, the sign-in is for one account, which needs API access turned on for the higher limits: if you belong to several, you choose which one ChatGPT works in. It can read data and run screens and rankings; it cannot open your projects or change anything in your account. It stops working if you leave that account, and until you accept updated Terms of Service. Disconnect it any time in Settings under Authorized apps. The tools the server offers are described on the MCP server page.

Example prompts

Shortlist metros for a plant

Shortlist US metros for an aerospace assembly plant: at least 500K people, manufacturing at least 8% of GDP, unemployment under 5.5% and industrial power under 10 cents per kWh. Rank the survivors on manufacturing share and unemployment, and list the sources with their retrieval dates.

Compare three metros on labor

Compare Huntsville, AL, Wichita, KS and Tulsa, OK on labor: unemployment rate, manufacturing employment growth, labor force growth and population. Make a table with each figure's vintage and source, and say which figures Cadastre computed.

Cheap power, growing manufacturing

Which metros have the lowest industrial electricity price and growing manufacturing employment? Show the ten best with the price, the employment growth and the years it covers.

Profile one metro

Give me Cadastre's profile of the Wichita, KS metro: population, GDP, manufacturing share, unemployment, power price and natural hazard risk, each with its source.

Common workflows

• Shortlist, then rank. Screen the metros on hard requirements first (the funnel says how many each one removes), then rank the survivors on weighted criteria. Ask the agent to show the funnel so you can see which requirement bites.
• Compare a shortlist. Give the agent three to five metros by name or CBSA code and the measures you care about; it reads them side by side and can put them in a table for a client.
• Look up before you ask. When a measure is ambiguous (growth over which years? which unemployment rate?), ask the agent to search the measures first and tell you which one it will use.
• Draft a client note. Have the agent write up a shortlist with every figure's source, vintage and retrieval date, and Cadastre's computed values labelled as such, then check the figures against the answers before you send it.

Tips

• Carry the attribution. Every answer lists the credit, licence and notices its sources require, and retrieved_at, when Cadastre last retrieved the data. Ask ChatGPT to keep them with the figures in anything it writes for you; the Terms of Service make it a condition, and Rights and attribution explains each source.
• Label Cadastre's computed values. A value marked computed_by_cadastre (such as a metro figure built from county figures) is Cadastre's, credited as "Source: Cadastre, computed from" the publisher's data.
• Check before you rely on it. An assistant can misread or misstate a figure or a credit. Compare what it tells you with the tool's own answer before you pass it on.
• Mind the rate limits. Without a sign-in, 30 requests a minute and 2,000 a day for each network address. Signed in, 60 requests a minute and 10,000 a day for each app you connect, and 50,000 a day for the account. Ask for a comparison or a screen in one go rather than metro by metro.
• Name metros precisely. "Wichita, KS" or its CBSA code (48620) avoids a wrong match.
• ChatGPT connects to Cadastre from OpenAI's servers, so without a sign-in it shares one allowance with other ChatGPT users: sign in (the sign-in address) for steady use.
• Custom MCP servers are added on ChatGPT on the web, subject to your workspace's permissions and security restrictions: see OpenAI's guide (opens in a new tab).

FAQ

What data does Cadastre have?

Public market, labor, demographic, infrastructure, hazard and health datasets from U.S. federal agencies, for the 393 U.S. metro areas, many also for counties and states. Datasets lists each one with its publisher, licence, period and measures. Only datasets whose terms allow it are served to agents.

Does it have listings or parcels?

No listings: Cadastre does not carry them. Parcel data is not available yet; the launch is metro-level.

Can it return boundaries or maps?

No. Answers carry names and codes (CBSA codes, FIPS), never boundaries, points or lines.

What does it cost?

The API and the MCP server are in beta and free while they are. The public datasets need no account; for higher limits, join now, then write to support@getcadastre.com to ask for API access, which may become part of a paid plan later.

What may I do with the answers?

Use them in your own work and for your clients, with the attribution carried, and not to rebuild Cadastre's data as a product of your own. See the Terms of Service.

Troubleshooting

Still stuck? Write to support@getcadastre.com.

Related agents

• Claude Code: Anthropic's agentic coding tool, in the terminal.
• Codex: OpenAI's coding agent: the Codex CLI, the IDE extension and the ChatGPT desktop app share one MCP configuration.
• Cursor: An AI code editor with an agent that can call MCP tools.
• GitHub Copilot in VS Code: Copilot's agent in Visual Studio Code.
• Claude: Anthropic's assistant on claude.ai and in the Claude desktop app, through a custom connector.
• Gemini CLI: Google's open-source AI agent for the terminal.
• OpenCode: An open-source AI coding agent for the terminal.

Sources

This guide's setup steps follow ChatGPT's own documentation. If a step no longer matches, theirs is current:

• OpenAI: Add custom MCP server (opens in a new tab)