Building a GeoAI agent that shows its work

The barrier to using geospatial data has always been about the technical expertise required to collect, manage, and analyze the data. But now, the agentic era of computation means that we can bring the vast capacities of geospatial analysis to a broader audience than ever before. To that end, we are excited to announce the geeViz MCP server.  

Based on a geospatial scripting library

In 2019, RedCastle Resources (RCR) developed geeViz: an open-source library designed to simplify complex geospatial processing and change detection in Google Earth Engine (GEE). For the last seven years, geeViz has been our daily driver for geospatial scripting, powering continental-scale remote sensing instances including the U.S. Forest Service’s Landscape Change Monitoring System (LCMS) and Tree Canopy Cover (TCC) datasets. These systems provide critical, peer-reviewed land cover identification and quantitative tree canopy predictions that are essential for long-range environmental assessment. 

GeeViz has its basis in sophisticated image compositing, cloud masking, and time-series change detection (complementing and integrating with other open-source libraries like geemap). Agentic intelligence gives us the ability to leverage these tools within a fundamentally new paradigm. 

Breaking through the GEE firewall 

Earth Engine is a live, highly secure, authenticated cloud platform. General-purpose Large Language Models (LLMs) tools can read or write abstract code about GEE, but they are blind to the mechanics of the system. LLMs cannot navigate the firewall to look at data stored in GEE or GCP, verify complex function signatures, or interact with a live session. To bridge the gap between LLMs and geospatial computation in GEE, RCR is launching the geeViz Model Context Protocol (MCP) Server.  

This cutting-edge implementation serves as a secure, real-time translation layer that allows AI coding assistants, like Claude or GitHub Copilot, to attach directly to a user's authenticated GEE session. By exposing the purpose-built tools in the geeViz library, we have given AI the tools it needs to inspect live metadata, and execute code within a persistent namespace. And importantly, the MCP is only the tool for writing the code, so the geospatial logic is preserved in the outputs: no black boxes here when it comes to your data outputs. The geeViz MCP allows developers and subject matter experts to orchestrate advanced geospatial workflows through conversational queries– a process we call “auditable vibe coding”.  

A geoAI tool that’s transparent, iterative, and open source

This advancement completely reimagines the geospatial lifecycle. Rather than just reading static code blocks, an AI agent can now:

• Access and remediate data: Pull unstructured data (like messy location logs), clean, and parse fields on the fly. 

• Explore and visualize: Generate live thumbnails, render interactive map views, and manage active cloud export tasks directly within the workspace.

• Iterate on the processing code: Because the agent operates transparently, developers can inspect the exact GEE JavaScript or Python code it generates, ensuring absolute reproducibility and trust.

We are proud to keep this foundational tool open source. The core geeViz MCP server tools are available today for developers to integrate into their daily IDE workflows and collaborative environments. For setup instructions and a deep dive into the open-source repository, visit our documentation link.