GISOwl MCP Server

Status: coming soon. Nothing on this page is available yet. Today an agent can call GISOwl's REST API with your workspace API key, the way the Python SDK does.

An MCP server will give an AI tool you already use, such as Claude, ChatGPT, Gemini, Cursor or a local model, the tools and sources Ask AI uses. It will run at GISOwl, so there will be nothing to install and nothing to keep running on your machine.

What an agent will be able to do

1. Find open data. Search the Data Hub catalogue and OpenStreetMap.
2. Query OpenStreetMap. Look up, rank, count and find the nearest features in GISOwl's PostGIS database, bounded by an area.
3. Build receipts. Buffer, intersect, join, filter and more, each step saved and rerunnable.
4. Publish and embed. Publish a map, then get its link and its embed code.

One server, for any agent

1. Your AI tool. "Which schools in Christchurch sit within 200 m of the 200-year flood extent? Put the map on our page." Asked in Claude Code, or in a local model.
2. GISOwl MCP server. Run by GISOwl. Tools in five groups: sources, data, receipts, maps, and Ask AI itself. The agent picks the ones it needs.
3. The tools it calls.
   • search_datahub → Christchurch City Council, 921 layers
   • add_datahub_layer → 144 schools, and the flood extent's 9,806 polygons
   • run_receipt → simplify 5 m, buffer 200 m, dissolve, intersect
   • publish_map → a public link and its embed code
4. The sources. The council's ArcGIS services and GISOwl's OpenStreetMap database. Each layer keeps its source, so it can be read again.
5. A live map and its receipt. 135 of 144 schools within 200 m, embedded in your page. It reruns when the council updates its flood model.

Planned setup

An agent will act with your API key for one workspace, inside that workspace only. Rotating the key cuts it off.