Skip to main content

Querying Overture Maps Data with an AI Agent

This example shows how to wrap Overture Maps Foundation open data as MCP tools on a Fused Canvas, so an AI agent can query buildings, places, and more by location or GERS ID without pulling raw geospatial data into the chat context. Try the chatbot or explore the Canvas.

Why AI Helps​

Overture exposes two different entry points into the data, and the right one depends on what the user is asking:

  • STAC catalog - best when the question starts from a place, address, bounding box, release, or theme.
  • Registry - best when the question starts from a specific GERS ID and needs feature history or an exact lookup.

The agent's job is to understand that intent and choose the right MCP tool automatically. That means the same chatbot can answer questions like "Are there any named buildings at Times Square, New York?", "When was GERS ID 0039aaa8-f0d9-45bb-90c5-d8a6781c4fbd first introduced?", and "How many building features are in the latest release?" without the user needing to know which data path to use.

How It's Built​

We build four UDFs, each scoped to one job, and publish them from a single Fused Canvas as an MCP server. Each UDF automatically becomes an HTTP endpoint, and the Canvas publishes an OpenAPI spec that any agent can load to discover available tools. Each tool returns only what the agent needs — bounded row counts, a fixed column set — while S3 reads, parquet scans, and geometry decoding stay inside Fused.

Fused Canvas showing the four Overture MCP tools

  • STAC catalog as the entry point — all tools read from catalog.json at stac.overturemaps.org, the authoritative index of releases, themes, and parquet file locations. No hardcoded S3 paths.
  • collections.parquet for bbox queries — a lightweight spatial index is queried first to find only the parquet partitions that intersect the requested bbox, before reading any data.
  • Binary search on the GERS registry — Overture publishes a registry manifest of 50 sorted files; a binary search on max_id pinpoints the single file containing the requested ID.
  • Address geocoding built in — a plain address string is geocoded via Nominatim and converted to a bbox automatically, so the agent never handles coordinates directly.

The Data Source​

Overture Maps Foundation publishes open releases of global geospatial data across two entry points: a STAC catalog for browsing releases, themes, and querying features by location; and a registry at s3://overturemaps-us-west-2/registry for looking up any feature by its GERS ID. The six themes available are:

ThemeDescription
buildingsBuilding footprints globally
placesPoints of interest, businesses, amenities
transportationRoad network segments
addressesAddress points
baseLand cover, water, land use
divisionsAdministrative boundaries

Creating MCP Tools​

The four UDFs below cover the full range of questions an agent might ask about Overture data — from catalog discovery to location queries to feature history.

overture_list_releases​

Lists every available release from catalog.json.

release        is_latest  schema_version  num_themes
2026-04-15.0   True                       6
2026-03-18.0   False      TBD             6

Browses themes, types, and feature counts for a given release.

theme           type               features
addresses       address            472,884,309
base            bathymetry              59,963
base            infrastructure     149,530,606
base            land                73,017,393
base            land_cover         123,302,114
base            land_use            54,159,685
base            water               64,677,485
buildings       building         2,542,597,608
buildings       building_part        3,905,964
divisions       division             4,597,236
divisions       division_area        1,074,405
divisions       division_boundary       87,804
places          place               75,495,994
transportation  connector          409,764,747
transportation  segment            344,140,478

overture_bbox_query​

Queries features by bounding box or address (geocoded automatically). Supports all themes, returns up to max_rows features (default 200).

address="Times Square, New York", theme="buildings":

id                                    type      subtype    class    name
54b78e32-a00c-4c8e-870b-a1279c6835ba  building
0039aaa8-f0d9-45bb-90c5-d8a6781c4fbd  building  education  college  Business and Liberal Arts Center
e26d5f68-186e-4162-8eba-fa04b4bd2fb9  building
f5518c97-3cd4-4faf-a1fe-2e30f72d8c02  building
ffe5fc63-8caf-4ef0-8339-68159f76354c  building

overture_gers_lookup​

Looks up a single feature by GERS ID using binary search on the registry manifest in catalog.json, then fetches the full geometry from the exact data file.

gers_id="0039aaa8-f0d9-45bb-90c5-d8a6781c4fbd":

id                                    name                              subtype    class    height  registry_first_seen  registry_last_seen  registry_last_changed
0039aaa8-f0d9-45bb-90c5-d8a6781c4fbd  Business and Liberal Arts Center  education  college  44.2    2025-06-25.0         2026-04-15.0        2025-06-25.0

See also​