--- name: brightdata-crunchbase description: Look up and discover companies on Crunchbase via Bright Data's pay-as-you-go Web Scraper API — pull a company's profile, founding date, HQ, headcount, industries, funding rounds, investors, acquisitions, patents, and key people; or discover companies by free-text keyword. Trigger whenever the user says "look up on Crunchbase", "research ", "who invested in ", "when was founded", "how much funding has raised", "find AI startups on Crunchbase", "discover fintech companies", "get Crunchbase data for …", or otherwise asks for Crunchbase-sourced company, funding, investor, founder, or startup data. Use this instead of web search whenever the user wants structured Crunchbase data — even if they don't explicitly say "Bright Data". --- # brightdata-crunchbase Thin wrapper around Bright Data's **Crunchbase Web Scraper API** (pay‑as‑you‑go, no minimum spend). Scripts hide the trigger → poll → download loop so callers stay task‑oriented. ## Prerequisites `BRIGHTDATA_API_TOKEN` must be set on a Bright Data account with the **Crunchbase Web Scraper API** enabled. Scripts exit with a clear error if the token is missing. ## Billing model — read this before running anything This skill calls Bright Data's Crunchbase Web Scraper API. Each **returned** record is billed per‑record on your Bright Data plan (pay‑as‑you‑go, **no minimum**, ~$0.001–$0.0015 per record). - **Lookups** (`lookup_companies.py`) bill for one record per input URL. Cheap and predictable — always prefer this when the user knows the company. - **Discoveries** (`discover_companies.py`) bill for every row returned. A broad keyword can return thousands of rows fast, so `--limit-per-input` is required. Before running a discovery, tell the user the expected max record count (= `len(keywords) * limit_per_input`) so they can ballpark the spend. - **Snapshot retention is 16 days.** Re‑downloading via `get_snapshot.py` is free. Don't re‑trigger for the same data inside that window. - `describe_fields.py`, `list_snapshots.py`, and `--status-only` on `get_snapshot.py` are free (metadata‑only, no scraping). ## Tasks Run from this skill directory; each prints one JSON object to stdout. - **Look up known companies** — `python scripts/lookup_companies.py --url https://www.crunchbase.com/organization/ [--url ...] [--fields name,url,founded_date,num_employees] [--no-wait]` Start here. One billable record per URL. - **Discover companies by keyword** — `python scripts/discover_companies.py --keyword "" [--keyword ...] --limit-per-input 25 [--fields ...] [--no-wait]` `--limit-per-input` is required. Start small (25) and expand only if needed. - **Re‑download an existing snapshot** — `python scripts/get_snapshot.py --snapshot-id s_xxx [--status-only]` Free within the 16‑day retention window. Use this instead of re‑triggering. - **List recent snapshots** — `python scripts/list_snapshots.py [--status ready] [--limit 20]` Free. Handy if the user lost a snapshot id. - **Describe the field schema** — `python scripts/describe_fields.py [--names-only] [--active-only]` Free. Authoritative source for what columns the scraper can return. ## Guidance - **Resolve the slug first.** The lookup path expects `https://www.crunchbase.com/organization/`. If the user gives a display name ("that sneaker startup"), either resolve the slug via web search or ask them for the URL. Don't guess — a wrong slug wastes a billable record. - **Cap discoveries.** If the user says "find me AI startups" without a number, default to `--limit-per-input 25` and tell them you capped it so they don't get surprised on their next Bright Data invoice. Increase only on explicit request. - **Default to the full record.** `lookup_companies.py` and `discover_companies.py` already return every available field when `--fields` is omitted — leave it omitted. Trimming saves nothing on the bill (billing is per-record, not per-field) and routinely strips out signals the caller didn't know they wanted (funding rounds, acquisitions, key people, patents). Only pass `--fields` when the user explicitly asks for a trimmed payload or context-window pressure forces it (e.g. `--fields name,url,founded_date,num_employees`). - **For long jobs, use `--no-wait`.** Record the `snapshot_id` in your response so the user can fetch later via `get_snapshot.py`. Large discoveries can take 10+ minutes. - **Nested fields, not separate endpoints.** Investors, funding rounds, acquisitions, key people, and patents all come back as nested structures on the company record. There are no separate endpoints for them. - **Prefer cached data.** Before triggering, ask whether the user already has a snapshot_id from a recent run — `get_snapshot.py` is free within 16 days of the original trigger. - **Don't dump raw JSON.** See the output contract below. ## Output contract For a single company, summarise the profile in 4–8 bullets: - Name (+ legal name if different) - Description (one sentence) - HQ city, country - Founded date - Industries (top 3) - Headcount range - Total funding raised + last round type/date - Key people (CEO, founders) if present - Website For a discovery result, summarise the top 5–10 hits in a compact table or bulleted list (name, URL, HQ, industries, headcount), then offer to drill into any specific company with `lookup_companies.py`. Always include the `snapshot_id` at the end so the user can re‑fetch for free. **Never dump the full JSON unless the user explicitly asks for "raw data".** ## References - `references/curls.md` — verified curl examples for every endpoint - `references/fields.md` — grouped field list (convenience; not authoritative) - `references/LIMITS.md` — billing, retention, auth, and failure modes ## Smoke test ```bash python scripts/smoke_test.py ``` Free, metadata‑only. Asserts the dataset metadata call works and that `name` and `url` are in the field schema. Prints `SKIP: BRIGHTDATA_API_TOKEN not set` if the token is missing, so unauthenticated CI doesn't fail.