Liquor License New Openings Tracker — TX, NY, FL Pre-Opening Leads

A multi-state liquor-license lead feed for hospitality vendors. A liquor-license application is the earliest public signal that a bar, restaurant, or liquor store is opening — typically 60–90 days before doors open, before the venue has a website, a phone number, or a POS contract. This actor pulls pending applications and newly issued licenses straight from official state registries and normalizes them into one clean, sales-ready schema.

• Texas (TABC) — active licenses + pending original applications (data.texas.gov)
• New York (SLA) — active licenses + pending applications (data.ny.gov)
• Florida (DBPR ABT) — licensed-file extract, active licenses (myfloridalicense.com)
• California (ABC) — scaffolded for v1.1 (see FAQ)

All sources are official, keyless open-data endpoints — no scraping, no logins, no keys.

Who buys these leads (who it's for)

• Beverage distributors & brand reps — reach a new bar before the first order is placed.
• POS, payments & reservation platforms — new restaurants buy their stack in the pre-opening window.
• Insurance agents (liquor liability, GL) — every new license needs coverage, often as a condition of the license.
• Food-service suppliers, equipment dealers, linen/hood-cleaning services — pre-opening procurement.
• Commercial real-estate & market analysts — track hospitality openings by city/county.
• AI sales agents — use it as an MCP tool to keep a CRM topped up with pre-opening leads.

Example input

{
  "states": ["TX", "NY"],
  "mode": "inventory",
  "feed": "pending",
  "maxResults": 200
}

Lead-feed setup: schedule the actor daily with "mode": "delta" and "feed": "all". The first run seeds a baseline (event_type: "inventory"); every later run emits only new_application, newly_issued, and status_change events — each one a fresh lead.

Other useful inputs: "city": "Austin", "county": "Bronx", "sinceDate": "2026-06-01".

sinceDate must be YYYY-MM-DD

sinceDate is a date predicate: it is spliced into the live Socrata $where clause on the TX/NY legs and compared against issue_date / application_date on the FL CSV leg. Both comparisons are on ISO date strings, so a value in any other format does not error — it silently returns the wrong set.

Accepted: 2026-01-15 and 2026-1-15 (single digits are padded). Leave it blank or omit it for no date filter — that is not an error and is not a default.

Anything else fails the run before any source is contacted, so nothing is billed. Slash forms are rejected on purpose rather than guessed: 05/06/2026 is genuinely ambiguous between US month/day and EU day/month order, and guessing produces a plausible wrong window instead of a visible error.

Measured against a 3-row fixture (2020-06-15, 2023-03-01, 2025-02-10) where the correct answer for a 2023-01-01 floor is 2 of 3:

You send	What it used to return	Why
01/15/2023	3 of 3 — the whole set, labelled as filtered	"01/…" sorts below every "20xx-…", so the predicate is true for everything
Jan 2023	0 of 3 — "nothing matched since Jan 2023"	"Jan…" sorts above every "2…", so the predicate is false for everything

Two more worth knowing, both now rejected or corrected: unpadded 2023-1-1 returned 1 of 3, silently dropping every month 01–09 (it is padded to 2023-01-01 before use); and 2023-13-45 — an impossible calendar date that satisfies a shape-only pattern check — returned 1 of 3, a plausible-looking wrong subset.

This matters most to AI agents calling the actor as an MCP tool: a date is exactly the value an agent formats however it likes, and neither failure mode is visible in the output.

Output fields

Each record is one license or application:

Field	Meaning
state	TX / NY / FL
feed	pending (application — pre-opening lead) or active (issued)
event_type	inventory, new_application, newly_issued, status_change
business_name	Trade name / DBA (falls back to owner/legal name)
owner_name	Owner / legal entity
premise_address, city, county, zip	Licensed premises location
license_number	License or application id
license_class_raw / license_class_label	State class code + decoded label
business_type	Decoded: bar, restaurant, package_store, manufacturer, wholesaler, club, other
status / previous_status	Current status (+ prior status on status_change events)
issue_date / application_date	Normalized YYYY-MM-DD
source, source_url	Upstream registry provenance

Use as an MCP tool

This actor is callable by AI agents (Claude, Cursor, etc.) via mcp.apify.com — the input and output schemas are fully described at field level, so an agent can query "new bar applications in San Antonio since June" and pipe the leads into a CRM without custom glue code.

FAQ

How early are pending-application leads? States require the license before alcohol service, and applications are filed while the venue is in build-out — typically 60–90 days pre-opening (varies by state and class).

Which feed should I use for lead generation? feed: "pending" for the earliest signal; feed: "all" + mode: "delta" on a daily schedule to also catch the moment a license is actually issued (newly_issued).

Is California included? Not yet — CA ABC publishes daily issued/new-application/status-change exports at a stable URL, but as a ZIP bundle whose layout couldn't be verified reliably at build time. The CA adapter is scaffolded and disabled; it ships in v1.1. Requesting "CA" today logs a warning and is skipped (never an error).

What does Florida cover? The FL leg reads the DBPR ABT licensed-file extract bd4006lic.csv (active beverage licenses; the sample profession file). Sibling extracts (bd4005/bd4002 — additional beverage series) and a pending signal are planned for v1.1. FL contributes to the active feed only.

How is business_type decoded? Per-state maps: TX TABC permit codes (MB→bar, RM→restaurant, P/Q→package_store, W/X→wholesaler, N→club, BB/BW/G→manufacturer…), NY SLA class descriptions (Restaurant, Tavern, Grocery/Liquor Store, Wholesale, Farm/Winery…), FL series patterns (…COP→bar, …SRX→restaurant, …APS/PS→package_store, 11xx→club). Unknown codes are kept raw and typed other — nothing is dropped.

Does an empty result cost anything? Pricing is per result record. A delta run where nothing changed emits 0 records and costs only the platform's minimal run overhead.

Is this legal to use? All data comes from official state open-data portals published for public reuse. Verify license details with the issuing authority before relying on them for compliance decisions; this is a lead-generation and monitoring tool, not legal advice.