New Business Leads — Verified Openings by State & County (SOS + Sales Tax)

New business leads from official state data, one row per business that just opened. Each opening is resolved across every register the state publishes — the Secretary of State charter (the entity exists) and, in Texas, the Comptroller sales-tax permit with its first-sales date (the business actually sells) — and carries the evidence that vouches for it, a three-valued opened_confidence, county, NAICS, entity type, and the trading address. Keyless, sourced from state open-data portals, no anti-bot.

Seven states, live-verified on every run: Texas, Colorado, Connecticut, New York, Oregon, Pennsylvania, Delaware — 18.6 million records in total.

It is also still the US Secretary-of-State business registry search and new-business monitor it has always been: search and new_businesses modes are unchanged, byte for byte, and Texas is now a seventh state in both.

The honesty boundary — read this before you buy leads

• An SOS charter is not proof of trading. It proves an entity was formed. Many chartered entities never open a door. A state that publishes only an SOS register (CO, CT, NY, OR, PA, DE) can therefore never produce an opening better than opened_confidence: low, and its opening_status can only read registered or planned. The row says so on registers_available.
• A sales-tax permit whose first-sales date has arrived IS proof of trading. Texas publishes the date each outlet first became liable to report sales tax. When that date is at or before today the row reads opening_status: trading; when it is still ahead — the Comptroller issues permits for outlets that open later; 6,663 such rows were live on 2026-09-26 — the row reads planned, which is the same "opening soon" signal Google-Maps scrapers sell, taken from the state's own record.
• Two registers agreeing = high. A charter and a first-sales date within 180 days of each other. Two registers that disagree by more than 180 days = medium (an existing business opening another outlet — see opening_kind). One register that proves trading = medium. One register that does not = low.
• A register that did not answer never lowers your row to a confident negative. registers_unavailable names it; franchise_record_on_file / sales_tax_permit_on_file stay null ("not checked") rather than false.

State coverage

Texas is the only state where two independent registers can be joined — on the Comptroller taxpayer number, an exact key shared by both rosters — which is why it is the only state that can reach high.

Scope and cost — read this first

states is required. There is no default and no hidden nationwide mode, so a run always answers about the states you actually named. A call with no states is rejected before the run starts — HTTP 400, no compute, no rows, nothing billed.

What a run costs. Billing is pay-per-result at $2.50 per 1,000 rows (about $0.25 per 100 rows), and paid Apify plans get a graduated discount of 20–70% off that. Only rows actually written to the dataset are billed — a run that fails a pre-flight check emits nothing and bills nothing. In openings mode one row is one business, so maxResults: 50 is at most 50 businesses ≈ $0.13.

maxResults is the hard cap. It bounds the whole run, across states, in every mode, and is applied exactly once by a single emission planner. In openings mode the cap is shared max-min fairly across the requested states (a TX + CO run at 50 gives each state 25 while both have rows), and within a state rows are ranked by opened_confidence then newest evidence. In openings mode a run that omits it caps itself at 200; in search / new_businesses a run that omits it behaves exactly as before (only maxPerState applies). Rows carry result_truncated: true when the cap bound.

Who it's for

• B2B sellers to brand-new businesses — payments, payroll, insurance, POS, web and marketing agencies, commercial real estate, business banking — prospecting weekly by county and NAICS.
• Lead-gen & RevOps teams that need "who just entered the market", not a registry dump.
• KYB / KYC / compliance teams verifying that an entity is registered (search).
• AI agents needing a "find / verify / monitor US businesses" tool.

Modes

openings — one row per business that just opened (new)

Every register behind the requested states is health-checked live, then queried for the window (daysBack, default prefill 30, or sinceDate). Records are merged into businesses on, strongest first: the Texas taxpayer number, the SOS file number, a register's own record id (this is what collapses Oregon's four associated-name rows), and finally normalised name + street address (or zip). Each business becomes one opening_signal row ranked by opened_confidence then newest evidence; the cap is shared fairly across states and the top rows per state are emitted. Texas rows are then enriched: for every emitted business the other Comptroller roster is looked up by taxpayer number (50 ids per request), so a charter-only row learns whether a permit exists, and a permit-only row learns its charter date, SOS number, entity type and standing.

{ "mode": "openings", "states": ["TX", "CO"], "daysBack": 30, "maxResults": 50 }

search — look up companies by name (unchanged)

{ "mode": "search", "states": ["CO", "CT", "NY"], "query": "Acme", "maxPerState": 50 }

new_businesses — raw recently formed entities, one row per registry record (unchanged)

{ "mode": "new_businesses", "states": ["TX", "NY", "PA"], "daysBack": 30, "maxPerState": 100 }

Delta / monitoring — monitor: true, any mode

Schedule the run and each run emits only entities (or, in openings mode, businesses) no previous run of the same scope has reported. The baseline lives in a named key-value store and is scoped to the mode, state set and query, so a narrow run cannot overwrite the baseline a wide run built. The first run of any new scope emits everything (delta_baseline_size: 0); every run after that emits only what is new. Records with no id are always emitted rather than silently dropped.

It tells you when a register did not answer

Every register is health-checked live before any result is returned — row-count floor, freshness (for the Texas rosters, whose newest dates are legitimately in the future, freshness is a floor on the trailing-30-day count), required columns, a closed-vocabulary check, and a negative control (a filter that must match nothing; if it matches something, the portal is ignoring the WHERE clause and every result would be an unfiltered slice of the registry).

A state that fails is excluded and named, on every single row: partial_results, states_ok / states_failed / states_failed_detail. In openings mode the disclosure is per register: registers_available, registers_checked, registers_unavailable. A Texas run in which the sales-tax roster is down still runs on the franchise roster — and every Texas row then says its confidence was computed without the permit register.

If no register can answer, the run fails loudly and bills nothing rather than returning a green, empty dataset that reads like a verified negative.

Date format — sinceDate

sinceDate must be YYYY-MM-DD. For example 2023-01-15. An unpadded month or day (2023-1-5) is accepted and padded; surrounding whitespace is trimmed. Leave it empty (or omit it) to use daysBack instead.

Anything else fails the run before any state portal is contacted, and bills nothing. The value is spliced into a live SoQL $where date literal against each register's own date column, so an unchecked format changes what question the portals are asked. Measured against data.colorado.gov on 2026-08-20:

Slash forms are refused rather than guessed at: 05/06/2025 is genuinely ambiguous between US month-first and EU day-first order.

Output fields

Every field below is declared in the dataset schema. Fields are nullable: null means the register does not publish it or it was not checked, never "we checked and there is none". false always means checked-and-negative.

Entity (all modes)

Registered agent

Address & geography

Texas registers (search / new_businesses / openings)

Opening signal (openings mode)

Oregon record shape (search / new_businesses)

Oregon publishes one row per associated name: record_type (

PRINCIPAL PLACE OF BUSINESS

Connecticut ownership flags (state self-certification)

woman_owned, veteran_owned, minority_owned, lgbtq_owned, disability_owned. null on every other state.

Run integrity — on every row

Use as an MCP tool

Exposed to AI agents via mcp.apify.com as a "find new businesses / search US business registries" tool — clean inputs (states, mode, window, cap) and structured output with explicit per-register status, so an agent can tell a verified opening from an unavailable source.

FAQ

Which states are covered? Texas, Colorado, Connecticut, New York, Oregon, Pennsylvania and Delaware, from each state's official open-data portal.

Why is a Colorado opening always low? Colorado publishes one register — the SOS entity roster. A charter proves the entity exists, not that it trades. The row tells you this on registers_available: co_sos.

Why can a Texas date be in the future? Texas allows a delayed effective date of formation, and the Comptroller issues sales-tax permits for outlets that will start selling later. Both are real, forward-looking signals; they read planned.

Is this a KYB API? It is a screening tool, not a determination. For a legal opinion on standing, pull the certificate from the state.

Why did a state return nothing? Check states_failed / registers_unavailable. If the state is not listed there, it answered and genuinely had no match. PA republishes roughly monthly, so a 7-day window can legitimately be empty.

Any anti-bot or captchas? No — this reads open-data APIs, not lookup pages.

Source

State open-data portals on Socrata: data.texas.gov (Comptroller of Public Accounts, two rosters), data.colorado.gov, data.ct.gov, data.ny.gov, data.oregon.gov, data.pa.gov, data.delaware.gov. Keyless. An optional free socrataAppToken raises rate limits. Texas code tables are transcribed from the Comptroller's published record layout.

Pricing (Pay Per Result)

Billed per row returned — one business per row in openings mode — with graduated discounts on paid Apify plans. A run that returns nothing costs nothing, and a run that cannot answer fails instead of billing.

Related actors

Other feeds from the same catalogue for lead-gen, RevOps and KYB teams working new and changing businesses. The Where both cover column lists only the places both actors actually serve.