Connecting CARTO to your agents

This quickstart connects your AI assistant to CARTO and builds your first map from a natural-language prompt — no code, no CLI. You'll use the CARTO MCP Server, which plugs CARTO into Claude, ChatGPT, Gemini Enterprise, Microsoft Copilot Studio, and other AI clients.

What you'll need:

• A CARTO account with a connected data warehouse (the built-in CARTO Data Warehouse works fine).

• An AI client that supports remote MCP servers — Claude is the quickest to set up.

Step 1: Get your MCP Server URL

In CARTO Workspace, go to Developers > Workflow API & MCP Server and copy your MCP Server URL. It looks like:

https://<region>.api.carto.com/mcp/<account_id>

If your organization uses single sign-on, copy the SSO login URL instead.

Step 2: Connect your AI client

Add the URL as a new MCP server in your client, then sign in to CARTO when prompted. Claude connects with just the URL — nothing else to configure.

• Claude (Claude.ai and Desktop)

• ChatGPT

• Microsoft Copilot Studio

• Google Gemini Enterprise

• All platforms →

Sign in over OAuth for the full experience — creating and editing maps, running Workflows, and admin. An API Access Token is a quick read-and-discovery alternative (browse, query, and search — no authoring).

Step 3: Ask your agent to build a map

Once connected, confirm the tools are live with a simple prompt:

List my CARTO connections and the tables in each.

Then build your first map. Paste this prompt and adjust the table name to one of your own (or use a Data Observatory dataset):

Explore my CARTO data, then create a Builder map from <your_connection>.<your_schema>.<your_table>. Pick a sensible geometry or spatial-index column, style it clearly, add a legend, and set the map to private. Show me the result and give me the link to open it in CARTO.

Your agent will explore the connection, inspect the table, compose the map, create it in CARTO Builder, and return a link. On hosts that render maps inline (Claude.ai, ChatGPT, Claude Desktop), it also shows the map right in the conversation.

Example: an agent builds a map from a prompt and renders it inline

Keep going

Try these next:

Add a histogram widget and a category filter to that map, then make it public.

Build a Workflow that filters this table to the last 30 days, spatially joins it to my regions table, aggregates by region, and saves the result. Then run it and summarize the output.

Enrich my stores table with population and income from the Data Observatory, then map the result.

Where to go next

• CARTO MCP Server — everything your agent can do, and how to connect each platform.

• MCP Tools Reference — the full tool catalog.

• CARTO Agent Skills — teach coding agents (Claude Code, Codex, Gemini CLI) the right CARTO patterns.

• CARTO CLI — scripting, CI/CD, and advanced automation.