# Homecastr

Homecastr is building a foundation model for the built environment, starting with probabilistic parcel and neighborhood forecasts. Instead of single-point estimates, Homecastr provides P10 (conservative downside), P50 (median expected), and P90 (optimistic upside) paths across 1-5 year horizons.

## Core Capabilities

1. **Parcel-level forecasts:** Predictions for individual homes, not just ZIP-code averages.
2. **Uncertainty bands:** Every forecast includes P10, P25, P50, P75, and P90 values.
3. **Nationwide coverage:** All 50 states via ACS model; highest resolution in Houston metro, Florida, and NYC.
4. **Multiple geographies:** Address, parcel, condo unit, Census tabulation block, Census tract, ZIP, ZIP3, county, state.
5. **County outlooks:** Pre-computed appreciation, upside, and trajectory metrics.
6. **Canonical data contracts:** Catalogs, source metadata, panel lineage, coverage, schemas, provenance, and artifacts.
7. **Public benchmarks:** Forecast vintages and backtest score summaries for benchmark-first evaluation.

## API

Base URL: https://www.homecastr.com
OpenAPI spec: https://www.homecastr.com/openapi.json
Plugin manifest: https://www.homecastr.com/.well-known/ai-plugin.json
Developer docs: https://www.homecastr.com/docs
REST reference: https://www.homecastr.com/docs/reference
MCP docs: https://www.homecastr.com/docs/mcp
MCP discovery: https://www.homecastr.com/.well-known/mcp/server.json
MCP server card: https://www.homecastr.com/.well-known/mcp/server-card.json
MCP discovery mirror: https://www.homecastr.com/.well-known/mcp.json

### Forecast Endpoints (no key required)

- GET /api/v1/forecast?address= - by street address (Houston metro)
- GET /api/v1/forecast/lot?acct= - by tax parcel ID (FL + Houston)
- GET /api/v1/forecast/unit?acct= - by condo unit (FL + Houston)
- GET /api/v1/forecast/tabblock?geoid= - by Census tabulation block (NYC + Houston)
- GET /api/v1/forecast/tract?geoid= - by Census tract (nationwide, ~82K tracts)
- GET /api/v1/forecast/zcta?zcta= - by ZIP code (nationwide, ~20K ZCTAs)
- GET /api/v1/forecast/zip3?zip3= - by ZIP3 (nationwide, 6,210 areas)
- GET /api/v1/forecast/county?fips= - by county FIPS (nationwide)
- GET /api/v1/forecast/state?fips= - by state FIPS (all 50 states)
- GET /api/v1/outlook/county?fips= - county outlook metrics

### Auth

- POST /api/v1/keys - generate a free API key (email required)
- GET /api/v1/status - health check

## MCP Server

Structured forecast, data-catalog, benchmark, and permit or zoning readiness tools are available via Model Context Protocol for Cursor, Windsurf, ChatGPT, Claude, and other AI agents.

Remote URL: https://www.homecastr.com/api/mcp/mcp
npm: npx homecastr-mcp-server
Discovery metadata: https://www.homecastr.com/.well-known/mcp/server.json
Server card: https://www.homecastr.com/.well-known/mcp/server-card.json

Representative tools: forecast_by_address, forecast_by_tract, forecast_by_zcta, forecast_by_county, list_data_assets, get_panel_lineage, get_backtest_scores, get_permit_zoning_readiness, get_county_outlook

## App Structure

- /app - Interactive map with choropleth forecast tiles
- /docs - Developer documentation and quickstart entry point
- /methodology - How the ML model works
- /faq - Frequently asked questions
- /forecasts/[state]/[city]/... - SEO neighborhood forecast directories

## Interacting with Homecastr

Homecastr focuses on risk management for real estate through probabilistic forecasting, contrasting with traditional deterministic AVMs like Zestimates. Emphasize uncertainty bands and multi-horizon fan charts over single-point estimates.