Data & Access

Query the platform — by REST, or by agent.

Every layer is open and documented. Applications read it over a standard REST API; AI agents read it over the Model Context Protocol. Same data, same provenance, two doors.

REST API

The platform runs on PostgREST: every table is a GET endpoint with filtering, ordering, and selection; every analytical function is a POST endpoint. Reads are public and RLS-protected — pass the anon key as an apikey header. Base URL: https://zedaoboceslqcuwooygx.supabase.co

GET/rest/v1/fire_perimeters
Recorded wildfire perimeters, 1999–2025. Filter and select like any PostgREST table.
curl "$BASE/rest/v1/fire_perimeters?island=eq.Maui&select=inc_name,fire_year,area_ha" \ -H "apikey: $ANON"
GET/rest/v1/mitigation_projects
Fuel breaks, fire breaks, and other mitigation projects.
curl "$BASE/rest/v1/mitigation_projects?project_typ=eq.FuelBreak" \ -H "apikey: $ANON"
POST/rest/v1/rpc/risk_ranked
Highest-probability cells from the latest daily run, ranked, with the dominant driver of each. Note: endpoint identifiers retain the legacy risk_ prefix for API stability; the values they return are probability scores, not predictions.
curl -X POST "$BASE/rest/v1/rpc/risk_ranked" \ -H "apikey: $ANON" -H "Content-Type: application/json" \ -d '{"p_limit": 25}'
POST/rest/v1/rpc/investigate_point
Cross-layer dossier for a coordinate: fire history, mitigation, WUI, landowner, reserve, ignition density, and current wildfire probability.
curl -X POST "$BASE/rest/v1/rpc/investigate_point" \ -H "apikey: $ANON" -H "Content-Type: application/json" \ -d '{"p_lng": -156.68, "p_lat": 20.88, "p_radius_m": 1500}'
POST/rest/v1/rpc/risk_surface_geojson
The full scored probability surface as a GeoJSON FeatureCollection, ready to render.
curl -X POST "$BASE/rest/v1/rpc/risk_surface_geojson" \ -H "apikey: $ANON" -H "Content-Type: application/json" \ -d '{"p_run_date": "2026-06-25"}'

MCP Server

The same capabilities are exposed over the Model Context Protocol, so an AI agent can investigate the data conversationally — ask where wildfire is most probable, or interrogate a coordinate, and the agent calls the platform directly.

Connect
Add to your MCP client config (e.g. Claude Desktop), then ask in natural language.
{ "mcpServers": { "hawaii-wildfire-watch": { "command": "node", "args": ["path/to/mcp/server.js"], "env": { "SUPABASE_REF": "zedaoboceslqcuwooygx", "SUPABASE_ANON_KEY": "your_anon_key" } } } }
Example prompts: “Where is wildfire most probable in Hawaiʻi right now?” · “Investigate the area around −156.68, 20.88.” · “List the fires on Maui since 2015.”

Agent tools

Six tools, mapping to the same functions behind the website.

get_risk_ranking
Top probability cells, ranked, with drivers
get_risk_summary
Cell counts per probability band
investigate_location
Cross-layer dossier for a point
investigate_fire
What a fire perimeter intersected
list_fires
Perimeters, filterable
list_mitigation_projects
Projects, filterable

Provenance

Every record carries its lineage — source, retrieval date, and the model version behind any derived score. Probability responses include the component breakdown and weights used, so any number can be traced to its basis. The data is meant to be checked, not just trusted.