PropDocket API

Why this design

Quickstart

All endpoints accept JSON POST, return JSON, and authenticate via a server-side API key in the x-api-key header. Keys are issued per organization on request — see the bottom of this page.

# Look up the flood zone for any Florida address (must include "FL")
curl -s -X POST https://api.propdocket.com/v1/get_property_flood \
  -H "x-api-key: $PROPDOCKET_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"address":"123 Main St, Anytown, FL 33000"}'

# Response (~480 B):
# {
#   "success": true,
#   "match_status": "exact",
#   "input": { "address": "123 Main St, Anytown, FL 33000" },
#   "location": { "latitude": ..., "longitude": ..., "matched_address": "..." },
#   "property": { "parcel_id": "..." },
#   "hazards": {
#     "flood": {
#       "payload": {
#         "zone": "X", "is_sfha": false, "static_bfe": null,
#         "vertical_datum": "", "dfirm_id": "12099C"
#       }
#     }
#   }
# }

Endpoints

All endpoints are POST. /v1/... base path. Sizes are typical responses for a Florida residential parcel.

Narrow — single-purpose, fast, cheap

Wide — full picture in one call

The fields parameter

All wide endpoints accept an optional fields: string[]. Pass dotted JSON paths or named groups. The response contains only the requested paths plus a thin envelope (success, match_status, request_id, input, location). Unknown paths are silently skipped.

When you pass fields, PropDocket also skips the upstream FEMA / USGS / ArcGIS / county-appraiser HTTP calls your slice doesn't need. Single-purpose responses are 5–8× faster than the wide-endpoint defaults.

Named groups

Example — explicit dotted paths

POST /v1/get_property_report
{
  "address": "123 Main St, Anytown, FL 33000",
  "fields": [
    "property.parcel_id",
    "property.owner_name",
    "hazards.flood.payload.zone",
    "hazards.wind.payload.categories.2.value.speed_mph"
  ]
}

include_* flags

Default to false on all endpoints. Pass true to opt back into the heavy data — typically only useful when debugging upstream parsing or building data exports.

Latency & size benchmarks

Real measurements from a Florida residential parcel. Numbers are warm-cache; cold-start adds 1–2 seconds for the runtime spin-up.

Use with Claude / AI agents

PropDocket ships an MCP (Model Context Protocol) server that exposes these endpoints as Claude tools — slim by default, with the same fields + include_* conventions. Agent-side responses are typically 12 KB or less without any prompt-engineering, well inside Claude's tool-result budget.

MCP endpoint

Install in Claude Code

claude mcp add --transport http propdocket https://propdocket-mcp-production.up.railway.app/mcp

Restart Claude Code after adding. The 12 propdocket tools (4 wide + 8 narrow) appear automatically.

Install in Claude Desktop

1. Open Settings → Connectors → Add custom connector
2. Name: propdocket
3. URL: https://propdocket-mcp-production.up.railway.app/mcp
4. Cmd+Q to fully quit Claude Desktop, then relaunch (window-close alone won't pick up new connectors)

Smoke test

Once connected, ask Claude:

• “What's the flood zone for any FL address?” — exercises get_property_flood
• “Verify this address and tell me the building department.” — exercises get_property_basics
• “What's the status of permit X in Pinellas County?” — exercises get_permit_status

The MCP is currently in private beta. The endpoint is open today (URL-obscured) while we finish user-side authentication. For team installs, deeper integrations, or production access, reach out via Request API Access.