AI for GIS: Connect Claude and ChatGPT to Your GIS Data With MCP

You can use AI for GIS by connecting Claude or ChatGPT to your GIS data through the Model Context Protocol (MCP). You sign in with your Geosys AI account, approve the access the assistant needs, and then ask it questions about your projects, datasets and analyses in plain language. The connector is available on request and is enabled per organization, so you need to ask us to switch it on for your team first. Once it is on, the assistant can only do what you allow and only within the projects you already have access to.

This guide explains what MCP is, what an assistant can do through the connector, how sign-in and scopes work, and where the limits are.

What is MCP?

The Model Context Protocol is an open standard that lets AI assistants connect to outside tools and data in a consistent way. Instead of copying data into a chat window, you add a connector, and the assistant can call the tools that connector exposes.

For GIS, that means your assistant can list your projects, look inside datasets, run an analysis and hand back the result, without you exporting files or pasting tables. Geosys AI provides that connector for both ChatGPT and Claude, so you can use ChatGPT for GIS as well as Claude. You can see the overview on the integrations page.

What can an assistant do with your data?

The connector offers more than 40 tools, grouped by what they do. Every tool respects the signed-in user's project permissions, so the assistant sees no more than you would.

Read

• List your projects and datasets.
• Search across projects, datasets and results in plain language.
• Fetch the details of a result, including tables, GeoJSON and charts.
• Check which uploads are waiting on a coordinate system confirmation.

Reading is the safest place to start. Ask "which datasets in this project have parcels?" or "summarize what the last analysis found."

Analyze

• Ask spatial questions and get answers back as tables, GeoJSON and charts.
• Start terrain and flood simulations and follow their progress.
• Run longer jobs in the background, with status you can check as they run.

Analyses, simulations, exports and uploads can take time, so they run as background jobs. The assistant polls for status and reports back when they finish.

Write

• Create layers in a project.
• Upload files inline, or pull a large file from an approved address.
• Supported uploads include shapefiles, GeoJSON, GeoPackage, CSV and XLSX, 3D models and CAD files.

Writing changes your project, so it is off unless you grant that access. Pulling files from a web address only works for addresses your administrator has approved.

Control the live map

If you have your Geosys AI dashboard open in a browser tab, the assistant can drive it. It can zoom to a dataset, toggle layers on and off, switch the basemap and apply filters. This needs a signed-in dashboard tab, and it is handy for walking a colleague through a finding: ask the assistant to "show the flood layer over the parcels" and watch the map change.

Bring your own model

If you prefer your assistant to do the analytical thinking, it can write GeoPandas code and run it in the Geosys AI sandbox, without using the Geosys AI analysis engine. The code is scanned before it runs. Network access, system calls and unsafe imports are blocked, and the run happens as a background job. Results are saved like any other analysis in your project, so you can open them later on the map.

How sign-in works

The connector uses OAuth, the same pattern you see when you connect any modern app to your account. The flow is:

1. In Claude or ChatGPT, you add the Geosys AI connector.
2. You are sent to a Geosys AI sign-in page and log in with your existing account.
3. You see the access being requested and approve it.
4. The assistant receives a short-lived access token, and a rotating refresh token keeps the connection working without you signing in again each time.

You never paste a password into the assistant. Access tokens are short-lived, and refresh tokens rotate whenever they are used.

Scopes: control what the assistant can do

Each connection is granted specific scopes, so you give an assistant only the access it needs:

• Read - browse projects, datasets and results.
• Analyze - run analyses and simulations.
• Write - create layers and upload files.
• Control - drive your open map dashboard live.

A sensible pattern is to start with read and analyze, and add write or control only when a task calls for it. Even with every scope granted, the assistant is still limited by your project permissions. A read-only project member cannot use the assistant to change a project.

The sandbox and other guardrails

Two guardrails are worth knowing about:

• Code is scanned and sandboxed. When the assistant writes analysis code, it runs in an isolated sandbox with unsafe imports, system calls and network access blocked.
• Usage is rate limited. Tool calls and code executions are limited per user, so a runaway assistant cannot flood the service.

For how access, sign-in and data control work across the whole platform, see Security and trust.

Limits to keep in mind

• Available on request. The connector is not on by default. It is enabled per organization, so contact us to get access.
• A signed-in user is required. The assistant acts as you. There is no anonymous access.
• Live map control needs an open dashboard tab. If no tab is open, the assistant can still read and analyze, but it cannot move the map.
• Large uploads have conditions. Pulling a file from an address only works for addresses your administrator has approved.
• Assistants can be wrong. Treat an answer as a starting point, check important results on the map, and keep a person in the loop for decisions.

Getting started

1. Request access for your organization.
2. Add the Geosys AI connector in Claude or ChatGPT.
3. Sign in and approve read and analyze access first.
4. Ask a question about a project you know well, so you can judge the answer.
5. Add write or control access only when you need them.

FAQs

Is the MCP connector available to everyone? No. It is available on request and enabled per organization. Contact us and we can switch it on for your team.

Which assistants does it work with? The connector works with Claude and ChatGPT.

Can the assistant see projects I do not have access to? No. Every action respects the signed-in user's project permissions, so the assistant sees only what you can see.

Can I stop the assistant from changing my data? Yes. Grant only the read and analyze scopes, and the assistant cannot create layers or upload files. Write access is a separate scope you have to approve.

Does the assistant need my password? No. You sign in through a Geosys AI login page using OAuth, and the assistant receives short-lived tokens instead.