This talk will demonstrate how to give AI tools access to your geodatabase, so colleagues can run geospatial analyses on your data using tools like Claude and ChatGPT. I work for Rebalance Earth, which requires large amounts of data to make decisions about investing in nature-based projects. These data are stored in a large PostGIS database. Frequently, less technical colleagues want to be able to run their own analyses using AI. However, we want to make sure that this is done in a safe and secure manner.

We initially developed a custom chatbot, however as the use of AI increased in the company it made sense for the app to migrate to an MCP server that any agent could access. I will start off with covering how I developed the initial toolbox to access our PostGIS database, getting the LLM to understand the structure and schemas of our geospatial data – including dealing with co-ordinate systems sensibly. Then I will talk about the move to an MCP server and the challenges faced with this. Topics will include: keeping the whole system safe and secure, ensuring that enough guidance was given to allow AI tools to produce the correct answer, dealing with large volumes of data without filling the AI model context, deploying using Azure Functions, and the quirks of developing MCP servers locally. I will hopefully be able to include live demos of using the MCP server.