US Building Permits Scraper — Contractor, Trade & Valuation

Building permits from 45 open municipal permit systems in 22 US states, normalised to one record shape and one trade vocabulary. 12.6 million permits of history sit behind these layers; a run returns the fresh ones.

This is a building permit lead feed, not a permit dump. Every record answers the four questions a contractor, roofer, solar installer or supplier actually pays for: who is doing the work, what trade it is, how much it is worth, and where.

Why it is not just "permit records"

A jurisdiction's permit type is not a trade. Columbus writes Roof, Siding, Windows, Doors, Nashville Building Residential - New, Anaheim Residential Remodel. The field workType maps all of them onto one small vocabulary (roofing, solar, hvac, electrical, plumbing, new-construction, remodel, demolition, signage, other), which is what lets someone who sells roofs filter for roofs in every city at once.

valuation never travels alone: valuationBasis says whether it is a declared job value or a construction cost. A $450 000 construction cost and a $2 685 job value are not the same number, and contractor totals never add them together silently — the portfolio row sets valuationBasisMixed instead.

Coverage — 45 jurisdictions, 22 states

State	Jurisdictions	State	Jurisdictions
AZ	3	MD	2
CA	5	MI	1
CO	4	NC	1
CT	1	NM	1
FL	4	NY	1
ID	1	OH	4
IL	2	OR	1
KS	1	PA	1
KY	1	TN	2
LA	1	TX	4
MA	1	WA	3

Every jurisdiction, its history size and the API it comes from:

City ID	Jurisdiction	State	Permits on record	API
us-tx-austin	City of Austin	TX	2 383 197	Socrata
us-ca-san-francisco	City and County of San Francisco	CA	1 297 749	Socrata
us-ny-nyc	New York City	NY	1 023 760	Socrata
us-ca-la-county	Los Angeles County (unincorporated + contract cities)	CA	928 761	ArcGIS
us-il-chicago	City of Chicago	IL	851 889	Socrata
us-oh-columbus	City of Columbus	OH	688 315	ArcGIS
us-fl-miami-dade	Miami-Dade County	FL	487 686	ArcGIS
us-wa-bellevue	City of Bellevue	WA	445 189	ArcGIS
us-la-new-orleans	City of New Orleans	LA	349 683	Socrata
us-co-boulder	City of Boulder	CO	337 797	ArcGIS
us-md-baltimore-city	Baltimore City	MD	295 869	ArcGIS
us-co-douglas	Douglas County	CO	286 727	ArcGIS
us-co-boulder-county	Boulder County (unincorporated)	CO	272 399	ArcGIS
us-md-bpds	Maryland Building Permit Data System	MD	257 533	ArcGIS
us-fl-miami	City of Miami	FL	236 389	ArcGIS
us-fl-alachua	Alachua County	FL	217 559	ArcGIS
us-oh-cleveland	City of Cleveland	OH	202 564	ArcGIS
us-wa-seattle	City of Seattle	WA	194 053	Socrata
us-ca-anaheim	City of Anaheim	CA	192 331	ArcGIS
us-nc-raleigh	City of Raleigh	NC	184 893	ArcGIS
us-oh-cincinnati	City of Cincinnati	OH	181 639	Socrata
us-ca-sacramento-county	Sacramento County	CA	159 396	ArcGIS
us-tx-san-antonio	City of San Antonio	TX	140 143	ArcGIS
us-wa-tacoma	City of Tacoma	WA	112 920	ArcGIS
us-az-coconino	Coconino County	AZ	111 645	ArcGIS
us-az-maricopa	Maricopa County	AZ	106 341	ArcGIS
us-ma-somerville	City of Somerville	MA	105 594	Socrata
us-fl-fort-pierce	City of Fort Pierce	FL	94 909	ArcGIS
us-nm-las-cruces	City of Las Cruces	NM	83 809	ArcGIS
us-tx-san-marcos	City of San Marcos	TX	51 811	ArcGIS
us-mi-detroit	City of Detroit	MI	47 998	ArcGIS
us-il-naperville	City of Naperville	IL	38 886	ArcGIS
us-ct-hartford	City of Hartford	CT	37 207	ArcGIS
us-tn-nashville	Metro Nashville / Davidson County	TN	29 661	ArcGIS
us-tn-memphis	City of Memphis / Shelby County	TN	28 404	ArcGIS
us-ky-louisville	Louisville Metro / Jefferson County	KY	22 639	ArcGIS
us-az-tempe	City of Tempe	AZ	20 744	ArcGIS
us-ks-overland-park	City of Overland Park	KS	14 197	ArcGIS
us-tx-kerrville	City of Kerrville	TX	11 098	ArcGIS
us-id-boise	City of Boise	ID	11 022	ArcGIS
us-ca-placentia	City of Placentia	CA	7 782	ArcGIS
us-oh-oxford	City of Oxford	OH	7 460	ArcGIS
us-co-aspen	City of Aspen	CO	6 553	ArcGIS
us-or-talent	City of Talent	OR	6 401	ArcGIS
us-pa-allentown	City of Allentown	PA	6 079	ArcGIS

37 layers are ArcGIS and 8 are Socrata. The Socrata half is what brought the largest systems into reach, because none of them publish an ArcGIS layer at all.

Verified live on 2026-11-02: all 45 jurisdictions answered, zero failures, 9 123 permits returned for a 7-day window (../../../docs/permits/measurement-notes.md).

What you get — one record per permit

{
  "source": "us-oh-columbus",
  "state": "OH",
  "permitId": "RSWDR2644234",
  "permitIdType": "accela-alt-id",
  "permitType": "1,2,3 Family",
  "permitSubtype": "Roof, Siding, Windows, Doors",
  "workType": "roofing",
  "status": "Permit Issued",
  "jurisdiction": "City of Columbus",
  "countyName": "Franklin",
  "countyFips": "39049",
  "issuedDate": "2026-10-31",
  "lastStatusDate": "2026-10-31",
  "valuation": 8000,
  "valuationBasis": "job-value",
  "situsAddress": {
    "street1": "3728 PENDLESTONE DR",
    "street2": null,
    "city": "Columbus",
    "state": "OH",
    "zip": "43230"
  },
  "parcelId": "600188600",
  "contractorName": "DTE ROOFING LLC",
  "scopeOfWork": "Additions, Alterations and Conversions - Residential",
  "sqft": 900,
  "units": 1,
  "completeness": "full",
  "omitted": [
    { "field": "ownerName", "reason": "not-published", "note": null },
    { "field": "appliedDate", "reason": "source-lacks-field", "note": "the layer publishes the issue date only" }
  ]
}

Group	Fields
Identity	permitId, permitIdType, permitType, permitSubtype
Trade	workType, scopeOfWork
Money	valuation, valuationBasis
Dates	appliedDate, issuedDate, finalizedDate, lastStatusDate
Status	status
Place	situsAddress (street1, street2, city, state, zip), parcelId, countyName, countyFips, jurisdiction, state
People	contractorName, contractorLicense, contractorPhone, ownerName
Size	sqft, units
Provenance	source, sourceApi, sourceLayerId, sourceUrl, scrapedAt, lookupType
Honesty	completeness, omitted[]

Two dataset views ship with it: permits (one row per permit) and portfolios (one row per contractor, free of charge).

omitted — why a field is empty

A missing value is never ambiguous. Every null carries an entry saying why:

Reason	Meaning
source-lacks-field	the jurisdiction publishes no such column
not-published	the column exists but is empty for this permit
derived	built from several columns (a glued address, a split ZIP)
zero-means-unknown	the source writes 0 where it means "not recorded"

So Nashville's status is null with source-lacks-field, not "cancelled". Anaheim publishes a contractor phone on under 9 % of rows. Both facts are in the dataset.

Inputs — cities, states, and dates

You choose which jurisdictions and which dates. All three selectors work together:

Input	What it does
cities	pick exact jurisdictions by ID (us-tx-austin, us-ny-nyc, …); empty means all 45
states	pick whole states (CA, TX, FL, …); combines with cities, a route listed twice is not fetched twice
lastNDays	window size in days, default 7
issuedAfter / issuedBefore	exact window, YYYY-MM-DD; either alone is honoured, and they win over lastNDays

{ "states": ["TX", "FL"], "lastNDays": 7, "maxResultsPerCity": 1000 }

{ "cities": ["us-ca-san-francisco", "us-il-chicago"], "issuedAfter": "2026-10-01", "issuedBefore": "2026-10-31" }

Four search modes run alongside the window:

Input	Question it answers
permitIds	what exactly was filed under this permit number
contractorNames	everything this contractor built, plus a free portfolio row
addresses	every permit ever filed on this address
mode: "new"	what was issued since the last run (default)

Window and limits

The ceiling of the window is today, and it is not cosmetic: several layers carry rows dated into the future, and a permit dated tomorrow has not been issued today — returning it on every run would bill the same permit twice.

maxResultsPerCity (default 1000) is per city, so adding cities never shrinks the others. A city that reaches it is marked truncated in the summary, which is how a cut-off stays distinguishable from a quiet city. Two hard ceilings sit above it: maxResultsPerCity cannot exceed 5000, and the whole run stops at 20 000 records.

The run summary lists, for every city, returned, inWindow and truncated — so a buyer can always tell "this suburb issued nothing" from "we cut you off at 1000".

Incremental mode

mode: "new" asks each layer for permits issued inside the window, newest first, and filters out everything an earlier run already pushed. The seen-set lives in the named store us-building-permits-state, one key per route, and the key deliberately does not contain the window — putting lastNDays in the key makes every new window look like a new search and re-charges the same permits.

Filtering happens before maxResultsPerCity is applied, so a page full of already-seen permits cannot starve a run of the new ones behind them.

Cost

$5 per 1000 permits — that is $0.005 per permit, and Apify adds its 20 % margin at checkout. One charged event (permit-scraped) per distinct permit pushed. A lookup that matches nothing is not charged, the contractor portfolio row is free, and a permit already returned by an earlier run is never charged twice.

A 7-day run across all 45 jurisdictions returns roughly 9 000 permits, which is about $45.

Honest limits

• Nine of the 45 feeds had nothing inside a 7-day window on 2026-11-02. That is the feed, not the actor: Boulder County's application feed lags six months, Aspen publishes closed permits only, Boise covers new housing and demolitions only. Each one still answers.
• Some columns are fees, not valuations (Bellevue TOTALBILLED, Coconino TotalFees, New Orleans totalfees, Cincinnati fee, Somerville application_amount) — they are not published as valuation and say so in omitted.
• A column named owner_contractor (Placentia) is published as neither, because the layer never says which of the two it is.
• New Orleans bldgarea holds only 0 or 4000, so it is a class flag, not an area, and is not published as sqft.
• Somerville publishes building permits, farmers-market licences and block-party licences in one table. The route carries an internal filter for department = 'ISD Building', so a permit lookup cannot return a licence that shares its digits.

Search terms

building permits, building permit scraper, construction permits, permit data, US building permits, permit leads, contractor leads, roofing permits, solar permits, HVAC permits, electrical permits, plumbing permits, new construction permits, remodel permits, demolition permits, permit valuation, job value, construction cost, issued permits, permit number lookup, contractor license, contractor name, owner name, parcel number, APN, property address, situs address, county FIPS, census tract, New York City permits, NYC DOB permits, Austin permits, Chicago permits, San Francisco permits, Los Angeles County permits, Seattle permits, Columbus Ohio permits, Cleveland permits, Cincinnati permits, Miami-Dade permits, Baltimore permits, Nashville permits, Memphis permits, Detroit permits, Sacramento permits, Anaheim permits, Raleigh permits, Tacoma permits, Bellevue permits, Boulder permits, Tempe permits, Maricopa County permits, Coconino County permits, Douglas County permits, Alachua County permits, lead generation, real estate data, construction intelligence, building activity, permit feed, daily permit feed, incremental scraping

Run locally

npm run stage -- us-building-permits
cd packages/actors/us-building-permits
mkdir -p storage/key_value_stores/default
# then write storage/key_value_stores/default/INPUT.json
APIFY_LOCAL_STORAGE_DIR=./storage node src/main.js

npm run stage inlines packages/core into src/core; without it the import ./core/index.js does not resolve outside the workspace.

The input directory is key_value_stores — plural. The SDK v3 creates it itself and silently ignores a hand-made key_value_store, so a run with a singular-name directory starts on the defaults and looks exactly like "nothing matched": zero records, zero misses, zero errors. That symptom means the input was not read; it never means the layers are empty.

Adding a jurisdiction

Adding a city is a row in src/routes.json, not a new parser, and a test checks the registry. The mapper enforces three rules: only mapped columns are requested (a missing column makes ArcGIS reject the whole query), 0 in a valuation column is "unknown", and every null gets an entry in omitted.

Tests

$node --test packages/actors/us-building-permits/test/*.test.js

408 tests. The fixtures are real rows from the layers, including the awkward ones: the zero valuation, the glued Anaheim address, the Nashville row with no status, the Somerville licence that shares digits with a permit.