About PropRaven

What we build

Every U.S. property, one lookup away. The canonical record covers 191.3M distinct parcels across all 50 states and DC, with 110.0M mapped locations; alongside the parcel spine sit 293M+ tax records, 180M+ building permits, 143M+ addresses, 27M+ resolved owners and 10M+ title transfers. Public figures are rounded down, never up — the data page carries the schema and the honest per-field fill rates, and coverage the per-state table.

There are eight ways in and one canonical spine underneath: a REST API, typed SDKs for Python, TypeScript, Go and dbt, an MCP server for Claude and AI agents, webhooks for sale, permit and owner-change events, a Snowflake Marketplace share, Databricks Delta Sharing, Parquet files on object storage for DuckDB, Spark and Polars, and the web and mobile app. Every door returns the same bytes. The Free plan is $0 — 1,000 lookups a month, no card required — and the paid ladder is on the pricing page.

How the data is made

Everyone who buys property data has been burned — a schema that drifted, a state that quietly shrank, channels that disagree — so trust here is a release process rather than a promise. In brief:

• Stable IDs. Every entity carries a deterministic identifier hashed from its canonicalized natural key: the same input produces the same ID every time, whitespace and case are tolerated, and IDs stay stable within a major version, so a warehouse can store them and join on them.
• Owner resolution. Owner names are clustered, classified by entity type (an obvious LLC, corporation or trust is never filed as an individual), and corporate entities with the same normalized name are merged into one canonical owner. Individuals and trusts are deliberately not merged, because first-name-only matches produce false positives; placeholder names such as “Current Owner” are dropped rather than served.
• Absentee classification. The property's situs address and the owner's mailing address are compared field by field; a mismatch is classified as out-of-state, out-of-county, local absentee or P.O. box, with confidence scaling with how many fields agree.
• Row-count floors, parity, epochs. A release can never quietly shrink — every table must clear the epoch it replaces or nothing ships. The warehouse share, the API and the app must return identical answers, checked daily and automatically. Every publish is an immutable, versioned epoch, and a bad one rolls back in one move.

The full write-up — confidence tiers, institutional-buyer classification, the known coverage gaps and what is deliberately excluded — is the methodology and versioning document, next to the data dictionary.

Audited nightly, in public

PropRaven audits its own data — every parcel, every field, against a published rule set — and ships the results as an attestation. Every customer-reported defect becomes a new rule under SLA. The audit page explains the pillars and the architecture, the findings are open to read, and the verify page lets anyone check a signed attestation offline.

Open code

The client libraries and the MCP server are public, and every published version carries a provenance attestation signed by GitHub Actions, so anyone installing PropRaven can prove the build came from a specific commit — the details are on the security page.

• · Python SDK: github.com/jdw2111/propraven-python · PyPI · propraven
• · TypeScript SDK: github.com/jdw2111/propraven-typescript · npm · @propraven/sdk
• · Go SDK: github.com/jdw2111/propraven-go · pkg.go.dev
• · MCP server: github.com/jdw2111/propraven-mcp · npm · @propraven/mcp
• · dbt package: github.com/jdw2111/propraven-dbt · dbt Hub

The company

PropRaven is built and operated by PropRaven LLC, based in Charlotte, North Carolina.

Data licenses and bulk questions go to sales@propraven.com; vulnerability reports to security@propraven.com. The terms of service, privacy policy and security practices are published in full, and shipped changes are logged on the changelog.