Case study · Built with Claude
Hurricane evacuation. FEMA flood. USDA hardiness. USFS wildfire. USGS seismic. EPA air-quality non-attainment. NOAA storm-surge. FEMA National Risk Index. All addressable by lat/lon, ZIP, or address through one HTTP API and one MCP server.
geographic-data-api is a generic, zone-type-agnostic backend for U.S. government geographic-zone data. Think of it as a single front door to the dozens of government agencies that publish zone-shaped datasets — FEMA, USGS, EPA, USDA, NOAA, USFS, state emergency management offices — with one consistent query interface.
Three products consume it today: EvacuationZones.com (Florida hurricane evacuation zone lookup), PlantGuideOnline.com (USDA-hardiness-aware gardening reference), and the new HardinessZones.com. The same API can drop into any project that needs to ask "what zones cover this point?".
LLMs that can call tools are at their best when the tools answer narrow, factual questions with citations. Government zone data is a perfect fit: the answers are unambiguous (this lat/lon is or isn't in flood zone AE) and every response carries the upstream agency and last-refreshed date so models can attribute correctly.
The MCP server exposes the API as seven model-facing tools:
list_zone_types() — discover what's available.describe_zone_type(slug) — schema, source, license, freshness.search_locations(q, kind?, state?) — autocomplete states / counties / towns / ZIPs.get_zones_at_point(lat, lon, types?) — point-in-polygon across every dataset.get_zones_by_zip(zip, types?) — fast denormalized lookup by ZIP.get_zones_by_address(address, types?) — geocode + PIP.get_plants_for_hardiness_zone(zone, type?) — the plant-guide use case.Install in any MCP-aware client with one line:
claude mcp add geographic-data \
--command python --args "-m app.mcp stdio" \
--env GDA_API_BASE_URL=https://api.zones.autonomous-agent.ai/docs.zone_lookups for the fast common case (ZIP→zone), PostGIS spatial query for arbitrary coordinates.From the architectural plan to the migration off Digital Ocean, the whole project was specified, scaffolded, and shipped through Claude Code over a handful of iterations. The plan, the schema, the ingesters, the MCP server, the deployment automation, and this very case-study page — all from the same harness.
That's the model we work from: AI as the disciplined, repeatable execution engine; humans as the operator, judge, and final checkpoint. If you have a backend you've outgrown — or a dataset you'd like to expose to LLMs — we'd like to talk.
Looking for search results instead? See our SEO case studies.