US Property Assessment Data Scraper — Assessed Value, Sale Price & Owner

Scrape US property assessment and sales data from the official statewide parcel layers that county assessors and state GIS offices publish themselves. No proxy, no API key, no login, no CAPTCHA. Thirteen states are measured and served, covering 45 542 174 parcels. Every number on this page comes from querying the layer's own returnCountOnly endpoint before the state was added — not from a vendor datasheet.

You pay for valuation, tax and sale depth, not for a name-and-address list. Of the 13 states, 10 return a valuation (land value, improvement value, total or assessed or market value), 5 return a last sale date, 4 a last sale price and 4 the tax year the value applies to — and where a source does not publish one of these, the row says so instead of leaving an unexplained blank.

Every state it serves, and the format each one returns

The point of this table is that you can see before you spend a cent whether your state and your fields are covered. Full means the layer carries that group for the whole state. — means the source does not publish the column at all, so no row from that state will ever contain it — and each such row says so explicitly in its omitted array.

State	Jurisdictions covered	Parcels	Returns owner	Returns situs address	Returns mailing address	Returns valuation	Returns county FIPS	Lookups that work
North Carolina (NC)	100 / 100 counties	5 938 804	yes	yes	yes	yes	yes	owner, address, parcel
New York (NY)	62 / 62 counties	5 510 061	yes (+ co-owner, owner type)	yes	yes	yes	yes	owner, address, parcel
California (CA)	58 / 58 counties	13 138 000	—	yes	—	—	yes	address, parcel
Colorado (CO)	44 / 64 counties	2 599 937	yes (+ co-owner)	yes	yes	assessed value	yes	owner, address, parcel
Arkansas (AR)	75 / 75 counties	2 149 448	yes	yes	—	yes	yes	owner, parcel
Connecticut (CT)	8 / 8 counties	1 282 833	yes (+ co-owner)	yes	yes	yes, plus tax year	yes	owner, parcel
Maryland (MD)	24 / 24 jurisdictions	2 395 737	—	yes	yes	yes	yes	address, parcel
Massachusetts (MA)	351 / 351 municipalities	2 559 749	yes	yes	yes	yes, plus tax year	—	owner, address, parcel
Utah (UT)	29 / 29 counties	1 600 062	—	yes	—	—	—	address, parcel
Virginia (VA)	county column absent	76 957	yes	—	yes	yes	—	owner, parcel
Washington (WA)	39 / 39 counties	3 321 859	—	yes	—	yes	38 of 39 counties	address, parcel
West Virginia (WV)	55 / 55 counties	1 394 081	yes (+ co-owner)	street line	street line	—	—	owner, address, parcel
Wisconsin (WI)	72 / 72 counties	3 574 646	yes (+ co-owner)	yes	yes	yes, plus tax year	yes	owner, parcel

The same table as rows and columns per state

State	Fields the layer actually fills	Fields the source lacks (declared per row)
NC	parcel, county name, FIPS, jurisdiction, owner, co-owner, owner type, land value, improvement value, total value, acreage, land use code, land use, year built, last sale date, situs, mailing	market value, assessed value, tax year, building sq ft, last sale price
NY	parcel, county, FIPS, jurisdiction, owner, co-owner, owner type, land value, improvement value, total value, market value, tax year, acreage, land use code, year built, building sq ft, situs, mailing	assessed value, land use text, sale date, sale price
CA	parcel, county, FIPS, situs address	owner (all owner fields), mailing address, every valuation field, acreage, year built, building sq ft, land use, sale date, sale price
CO	parcel, county, FIPS, owner, co-owner, total value, assessed value, acreage, land use code, land use, sale date, sale price, situs, mailing	jurisdiction, owner type, land value, improvement value, market value, tax year, year built, building sq ft
AR	parcel, county, FIPS, owner, land value, improvement value, total value, assessed value, land use code, situs	jurisdiction, co-owner, owner type, mailing address, market value, tax year, acreage, year built, sale date, sale price
CT	parcel, FIPS, jurisdiction, owner, co-owner, land value, improvement value, total value, assessed value, tax year, acreage, land use code, land use, year built, building sq ft, sale date, sale price, situs, mailing	county name, owner type, market value
MD	parcel, FIPS, land value, improvement value, total value, acreage, land use code, land use, year built, building sq ft, sale date, sale price, situs, mailing	county name, jurisdiction, owner (all owner fields), market value, tax year
MA	parcel, municipality, owner, land value, improvement value, assessed value, tax year, acreage, land use code, land use, year built, building sq ft, sale date, sale price, situs, mailing	county name, county FIPS, co-owner, owner type, market value
UT	parcel, county, owner type, situs address	FIPS, jurisdiction, owner name, co-owner, mailing address, every valuation field, acreage, land use, year built, building sq ft, sale date, sale price
VA	parcel, owner, land value, improvement value, total value, land use code, land use, mailing address	county name, FIPS, jurisdiction, co-owner, owner type, situs address, market value, assessed value, tax year, acreage, year built, building sq ft, sale date, sale price
WA	parcel, FIPS, land value, improvement value, land use code, situs address	jurisdiction, owner (all owner fields), mailing address, market value, assessed value, tax year, acreage, year built, building sq ft, sale date, sale price
WV	parcel, owner, co-owner, acreage, situs street line, mailing street line	county name, FIPS, jurisdiction, owner type, every valuation field, land use, year built, building sq ft, sale date, sale price
WI	parcel, county, FIPS, owner, co-owner, land value, improvement value, assessed value, market value, tax year, acreage, land use code, situs, mailing street line	jurisdiction, owner type, land use text, year built, building sq ft, sale date, sale price

California is the largest layer here (13.1 M parcels) and the sparsest in columns: it publishes geometry-linked parcel identity and address, and nothing about the owner. That is a property of the source, not a mapping gap — and the run summary reports it as such.

Which states actually return value and sale prices

This is the part most property datasets obfuscate, so it is on one screen.

State	Value fields returned	Tax year	Last sale	Physical
NC	land, improvement, total	—	date only	acreage, year built, land use
NY	land, improvement, total, market	yes	—	acreage, year built, sq ft, land use
CA	none	—	—	—
CO	total, assessed	—	date + price	acreage, land use
AR	land, improvement, total, assessed	—	—	land use
CT	land, improvement, total, assessed	yes	date + price	acreage, year built, sq ft, land use
MD	land, improvement, total	—	date + price	acreage, year built, sq ft, land use
MA	land, improvement, assessed	yes	date + price	acreage, year built, sq ft, land use
UT	none	—	—	—
VA	land, improvement, total	—	—	land use
WA	land, improvement	—	—	—
WV	none	—	—	acreage
WI	land, improvement, assessed, market	yes	—	acreage, land use

A dash is a fact about the source, and the same fact is repeated in the omitted array of every row from that state. Nothing here is inferred to fill a column.

The omitted array is the reason to trust a blank cell

Every canonical field exists on every record. When one is null, it is never silent — the row carries omitted, an array naming the field and the reason:

• source-lacks-field — the layer has no such column. No record from this state will ever have it.
• not-published — the column exists, but this row's value was blank.
• derived — the number was computed rather than read, and it says so (New York's improvement value; North Carolina's glued addresses).
• not-covered — the jurisdiction is outside the layer.
• lookup-miss — you searched, and nothing matched.
• filtered — your own input excluded it.

So lastSaleDate: null with { "field": "lastSaleDate", "reason": "source-lacks-field" } means "this county does not publish sale dates", not "this house never sold". Those are different statements and only one of them is true. A property data scraper that quietly leaves cells empty forces you to guess which. This one does not.

Input parameters

Parameter	Type	Default	What it does
ownerNames	string list	—	Owner-name search, one name per row, minimum 3 characters. Partial match: SILVER PINES finds SILVER PINES PROPERTIES LLC.
addresses	string list	—	Situs address search, one per row, minimum 4 characters, with a street-only fallback. 609 MAIN ST works better than 609 Main Street, Raleigh, NC 27601.
parcelIds	string list	—	Parcel number as the county writes it. 1234-56-789, 123456789 and 1234 56 789 are all tried. New York wants the PRINT_KEY value.
states	multi-select	all	Which of the 13 states to query. A state measured as broken is an error naming the measurement, not a silent empty dataset.
bulk	boolean	false	Ignore lookup lists and sample rows straight from the layer, to inspect a source before paying for lookups.
bulkCount	integer	50	Rows per state in bulk mode (max 1000). The layer's own maximum is respected and truncation is reported.
maxRecords	integer	50	Hard cap for the whole run across all states — this is also your cost cap (max 1000).
maxRecordsPerLookup	integer	10	How many parcels one owner name or one address may return. Stops a surname like SMITH from eating the budget.
ownerPortfolio	boolean	true	After every owner search, push one free owner_portfolio row: parcels returned, counties, value totals with the count of records behind each total, acreage, absentee count. Never billed.
includeRaw	boolean	false	Attach the untouched source attributes to each record under raw, for columns outside the mapped set.
maxConcurrency	integer	2	Parallel lookups. The layers are public and unauthenticated; 2 is polite and fast enough.
requestTimeoutSecs	integer	45	Per-request timeout. New York answers in 6–8 seconds from the platform, so do not go below 20.
chargePerRecord	boolean	true	Emits one billing event per distinct record pushed. Missed lookups are never charged.

Example — lookups across two states:

{
  "ownerNames": ["SILVER PINES PROPERTIES LLC"],
  "addresses": ["105 E MAIN ST"],
  "parcelIds": ["010100000100001000100000"],
  "states": ["NC", "NY"],
  "maxRecords": 200
}

Example — sample a state without spending on a lookup list:

{ "bulk": true, "bulkCount": 100, "states": ["NC"] }

Output

Two row shapes share the dataset. Parcel rows are one per parcel, identical across all 13 states; the free owner_portfolio row described below is the only other rowType. Filter on rowType if you feed the dataset straight into a pipeline.

Group	Fields
Identity	state, countyName, countyFips, jurisdiction, parcelId, parcelIdType
Owner	ownerName, ownerName2, ownerType
Property address	situsAddress (street1, street2, city, state, zip) and ownerMailingAddress
Valuation	landValue, improvementValue, totalValue, assessedValue, marketValue, valuationType, taxYear
Physical	acreage, buildingSqFt, yearBuilt, landUseCode, landUse
Sale	lastSaleDate, lastSalePrice
Provenance	source, sourceUrl, lookupType, scrapedAt, raw
Honesty	completeness, omitted

The free owner_portfolio row

A list of parcels does not tell you whether you found a landlord or a resident. So every owner search also emits one summary row of rowType: "owner_portfolio", and that row is never charged:

{
  "rowType": "owner_portfolio",
  "ownerQuery": "GLOVER FARMS OF CAROLINA LLC",
  "state": "NC",
  "parcelCountReturned": 10,
  "totalMatches": 16,
  "truncated": true,
  "counties": { "Northampton": 10 },
  "countiesCount": 1,
  "ownerNames": ["GLOVER FARMS OF CAROLINA LLC"],
  "valueTotal": 505779,
  "valueRecords": 10,
  "valueBasis": ["totalValue"],
  "lastSalePriceTotal": null,
  "lastSalePriceRecords": 0,
  "acreageTotal": 1170.55,
  "acreageRecords": 10,
  "absenteeCount": 0,
  "absenteeComparableRecords": 10,
  "totalsScope": "all_returned_records",
  "note": null
}

Three things make it honest rather than decorative. Every total carries the number of records that actually contributed to it, and totalsScope drops to records_that_publish_the_field the moment the state does not fill that field on every row — so a portfolio total is never silently divided by the wrong denominator. totalMatches comes from the layer's own returnCountOnly count on the same where clause, so you can see how much of the portfolio you were truncated out of. And absenteeCount only counts a parcel when the record publishes both the mailing state and the situs state; if it does not, the count is null with a note explaining why, because "0 absentee owners" and "I cannot tell" are different answers.

valueTotal follows the state, not a single field: it sums the best valuation each record publishes — assessedValue first, then marketValue, then totalValue — and valueBasis names which one it used. That is why North Carolina, which publishes only total value, still returns a portfolio figure instead of null.

States are inferred from the addresses you give

Leave states empty and the actor reads the state out of each address, so a Florida address queries Florida instead of sweeping all 13 states. It understands Lakeland, FL 33813, WI 546428226 with a glued ZIP+4, and the spelled-out Ohio. When every address resolves, the run narrows to those states and says so in the log; when only some do, it warns and queries everything rather than silently dropping the addresses it could not read.

A SUMMARY record lands in the key-value store with a per-state breakdown: the coverage statement, the layer's live row count and drift since it was probed, lookup hits and misses, bulk-slice stats, fieldAudit (which mapped columns the layer actually replied with and which were missing) and omissions (for each null field, how many records lacked it and why). If a source changes its schema under you, fieldAudit is where you see it, in the run where it happened.

How matching behaves

• Owner — LIKE '%value%' on the owner column. The column holds the name as the county recorded it, so try the surname first, then the full entity name.
• Address — LIKE on the situs street line with a street-only fallback, because layers store the county's own formatting and a full address with city and ZIP often fails to match it.
• Parcel — exact match, tried three times: verbatim, compact, compact without dashes.
• Deduplication happens before billing, so overlapping lookups never charge twice for the same parcel. A lookup that matches nothing costs nothing.

What it is not

• Not a nationwide scraper. Thirteen measured states out of fifty. The registry is data, so adding a state means adding an entry — but each entry needs a real probe first, and ten candidates were probed and rejected because their layers refuse queries or lack owner and address columns.
• Not a deeds or liens database. This is the assessor's roll: ownership, value, physical description. Transfers appear only where the source publishes them — you get the last sale, not a sale history, because the layers carry a current snapshot rather than a transaction ledger.
• Not a uniform field set. Four states publish the tax year, five publish a sale date, two publish market value. The table above is the contract; a field absent from it is absent from every row of that state, with the reason recorded in omitted.
• Not a tax-bill calculator. taxYear tells you which assessment year the value belongs to. The millage rate is set by each jurisdiction separately and is not in these layers.
• Not a guarantee of a situs address in every county. Address completeness is county-dependent and it was measured, not assumed: in North Carolina only 25 of 100 counties publish a situs city and 90 of 100 publish a situs street.
• Not a source of exact coordinates. Geometry is not returned; the parcel number and county FIPS join to any public parcel geometry you already have.
• Not free of the source's own lag. These are annual assessment rolls. A sale recorded last month is not in them yet.

Cost and speed

$5 per 1 000 records — $0.005 per record, and the Store adds its 20 % margin, so you pay $6 per 1 000. Plus one apify-actor-start event per run at $0.00005. You are billed per parcel record pushed to the dataset, not per request and not per lookup, and maxRecords is therefore a real spend cap. The owner_portfolio summary row is free, and so is every lookup that matches nothing — a run that finds no parcels costs only the start event. chargePerRecord: false disables billing entirely.

Most layers answer a query in about 2 seconds. New York takes 6–8 seconds, which is why the timeout defaults to 45. Concurrency defaults to 2 because the services are public and unauthenticated, and hammering them is how you get throttled.

Keywords this actor is built for

property assessment data scraper · property tax records by owner name · county assessor assessed value lookup · US parcel data with sale price · land and improvement value data · property owner mailing address list · absentee owner list · acreage and year built data · real estate lead generation by owner · situs address and parcel number search.

Frequently asked

Does it need a proxy? No. These are public ArcGIS and REST services, and the actor talks to them directly — which is also why it is fast and why no proxy cost is baked into the price.

What happens to states whose layer is broken? They are rejected at probe time and recorded with the exact error. Asking for one is a run error that names the measurement, so you never mistake "the source is dead" for "there is no such owner".

Can I get the raw columns? Yes — set includeRaw: true and each record carries the untouched source attributes under raw, alongside the normalised fields.

Do I get a tax history or a sale history? No, and neither does a scraper that claims to: these layers publish the current roll — the latest assessed value, the latest sale — not a yearly ledger. What you get instead is the year the value applies to (taxYear, where the source publishes it), so you can tell a 2023 assessment from a 2025 one, and omitted tells you when a state does not say.

How do I find absentee owners? Search the owner name across a state and read absenteeCount on the free portfolio row: it counts parcels whose recorded mailing state differs from the parcel's own state. It is null — not 0 — where the source publishes no mailing state, because a missing column is not evidence of an owner who lives on site.