Agentic GIS in Snowflake with Cortex Code and CARTO

A couple of weeks ago we introduced CARTO for Agents, making CARTO the first GIS platform fully designed for the Agentic Enterprise. With an extended CARTO CLI, a public catalog of CARTO Agent Skills, and a richer MCP Server, any coding agent your team uses can now operate CARTO end-to-end.

CARTO and Snowflake have been close partners for years. CARTO runs natively on Snowflake, with every spatial workflow, map, and application reading and writing your Snowflake data in place. No intermediate storage, no extra data governance to manage. In today's agentic world, we also wanted CARTO to integrate cleanly with Snowflake's AI surface from day one, just as it does with the rest of the platform.

Snowflake has been moving fast on its own agentic agenda. Cortex Code is Snowflake's native AI coding agent, built to operate inside your Snowflake environment with full awareness of your roles and grants, your schemas, warehouses, and governance posture. Snowflake Intelligence brings that same power to business users, letting teams interact with their Snowflake data through natural language.

CARTO for Agents pairs naturally with both, extending our Cortex AI integration all the way up to the coding-agent layer. In this post, we walk through how you can use Cortex Code with CARTO to build interactive maps, workflows, and spatial applications, and how CARTO's MCP Server brings auditable and traceable geospatial analysis to Snowflake Intelligence.

What’s CARTO for Agents?

CARTO for Agents is the product surface that lets any coding agent operate the CARTO platform on behalf of your team. It has three main pieces:

• CARTO CLI exposes nearly every CARTO operation as a script-friendly command: connecting to your data warehouse, importing data, creating CARTO Builder maps, designing CARTO Workflows, developing spatial applications, managing your assets and analyzing usage, and more.
• CARTO Agent Skills is a catalog of skill playbooks that teach coding agents how to drive CARTO correctly: which CLI flags to use, which SQL dialect to write, which patterns work for common spatial analyses (hotspot analysis, territory planning, composite scoring, trade-area analysis, and more).
• CARTO MCP Server exposes built-in CARTO tools and any workflow your team has published as MCP tools, ready for any MCP-compatible agent or agentic platform to call.

These three pieces are designed to work together. Your coding agent installs the Skills, calls the CLI to operate the platform, and reaches the MCP Server to render interactive maps or run the analytical procedures your teams have built to be deterministic.

Agentic GIS with Cortex Code

Cortex Code's local CLI bridges your development environment with your Snowflake account, operating inside your perimeter and inheriting your governance as it works. That's exactly the environment CARTO for Agents was designed to slot into.

Two installation steps and Cortex Code becomes a fluent CARTO operator:

npm install -g @carto/carto-cli

npx skills add CartoDB/agent-skills

The CARTO CLI gives Cortex Code the hands; the CARTO Agent Skills give it the judgment. From there, Snowflake and CARTO users can work with CARTO end-to-end all driven through Cortex Code, including:

• Builder maps. Cortex Code uses the carto-create-builder-maps skill to compose a CARTO Builder map from a natural-language request, including color palettes, layers, widgets, popups, validation, and publish. The map is a real, governed asset in your CARTO organization, shareable across the team.
• Workflows. Cortex Code uses the carto-create-workflow skill, and the use-case catalog: carto-hotspot-analysis, carto-territory-planning, carto-composite-scoring, carto-trade-area-analysis, and others, to design a spatial workflow against your Snowflake data, validate it against the warehouse, and ship a versioned, auditable analytical procedure your team can run on demand or on schedule.
• Applications. Cortex Code uses the carto-develop-app skill to scaffold a working CARTO + deck.gl application, with scoped API tokens minted via the CARTO CLI's credentials commands, the right framework, and Snowflake as the data source, running locally in seconds.

In all three flows, accessing the data and the spatial computation itself happens inside Snowflake. That's the core foundation CARTO offers, and it's what sets us apart from legacy GIS tools and geospatial platforms. CARTO is cloud-native in the strict sense. Every CARTO Workflow compiles to Snowflake-native SQL and runs in your warehouse; every map query the application fires executes inside Snowflake.

The governance you've already configured stays in force throughout the loop, with neither CARTO nor Cortex Code needing a parallel access layer. Speed at the surface, Snowflake governance underneath.

Bonus: CARTO Workflows as MCP tools for Snowflake Intelligence

Cortex Code serves data scientists and developers building on Snowflake. Snowflake Intelligence serves everyone else, such as the business users who want to interact with their Snowflake data through natural language without ever opening an IDE. Snowflake Intelligence is powered by Cortex Agents, which now support MCP connectors. Register any MCP-compatible server as a connector and its tools show up in the conversation alongside Snowflake's own.

This opens a clean integration path for advanced spatial work. A geospatial analyst on your team can design a CARTO Workflow, say, a site-selection scoring pipeline or a trade-area enrichment, using Cortex Code and the CARTO use-case skills. The same analyst then publishes the workflow as an MCP tool with carto workflows mcp publish, and the Snowflake admin registers the CARTO MCP Server as a connector in Snowflake Intelligence.

From that point on, any business user in Snowflake Intelligence can invoke the published spatial workflow in natural language. The agent calls the procedure your analyst built, validated, and signed off on, not a one-off SQL query or a model crafted on-the-fly. The flexibility of natural language meets the reliability of a vetted CARTO Workflow, with all of it running inside Snowflake.

Agentic GIS with Snowflake and CARTO

If your team builds on Snowflake, Agentic GIS is one Cortex Code session away. Install the CARTO CLI and Agent Skills, connect Cortex Code to the CARTO MCP Server, and start creating maps, workflows, and spatial apps with the assistance of an AI Agent governed by Snowflake. documentation to get started, or talk to the CARTO team.

You can also meet us at Snowflake Summit, June 1–4. Come see this in action and talk to the CARTO team about your use cases and Snowflake setup.