google-finance-scraper

Scrapes Google Finance — instrument quotes, quarterly financials, analyst ratings and price targets, OHLCV bars, the market index and sector strips, currency, crypto and futures, the day's gainers and losers, and the earnings calendar — with Playwright, Selenium, Puppeteer, or the 2Captcha Scraping Browser API over CDP. JSON and CSV out.

Seven modes, all reading the page's own server-rendered payload. No key, no proxy, no account.

---

What you actually need

No key, no account, no browser infrastructure — and no proxy, from a region Google serves Finance in.

That last clause is not hedging. A third-party audit hit it on 2026-09-24, and it was then measured directly through pinned residential exits — ten countries, one request each:

exit	result
RU	HTTP 403, refused (4 of 4: three quote fetches and the market page)
US, DE, FR, HU, PL, CZ, AT	HTTP 200, served

What a refused region gets:

403. That's an error.
Google Finance is currently not supported in your region.

--market does not get you past that, and the distinction matters because the flag looks like it should. Verified from the refused exit: ?gl=US&hl=en answered 403 with the same message, on both the quote route and the market page. gl selects WHICH market's data you are served; the regional gate decides whether you are served at all, and it is checked first.

From a refused region you need an exit in a supported one — a --proxy, or the country- segment of a Scraping Browser endpoint. From a supported region you need nothing.

The scraper now recognises that page as region_unavailable and reports exit 3 with advice naming the country axis, rather than the exit 4 it used to report, which means "the page loaded and the market is empty".

Everything below was measured from one datacentre address (Hetzner, Helsinki, AS24940) on 2026-09-22, with no credential of any kind:

command	result
--mode markets --market US	46 rows, every one priced
--mode movers --market US	11 rows across three lists
--mode quote --symbols GOOGL:NASDAQ,BMW:ETR,EUR-USD,BTC-USD	4 rows in 8.3 s
--mode financials --symbols GOOGL:NASDAQ	90 quarters, back to 2004
--mode analysts --symbols GOOGL:NASDAQ	44 analyst actions + the consensus
--mode chart --symbols GOOGL:NASDAQ	99 bars — 79 intraday, 20 daily
--mode earnings --market US	5 upcoming announcements

Within a supported region, the site is unusually open, and the reason is worth stating plainly because it changes what the paid products are for: which market you read is a query parameter, not a property of your exit address. --market US from a Finnish address returns NASDAQ movers and CBOE sector indices; --market DE returns ETR and STOXX; with no --market at all you get whatever market your exit IP sits in. Measured, one fetch each, same address.

So one usual reason to buy an exit — "I need to see the American market" — does not apply. The other one does: if your own region is refused, an exit is the only way in. What the 2Captcha products buy:

• proxies — access from a refused region, and volume. Spreading a large run across exits is what keeps any one address from being scored.
• the Scraping Browser API — no local browser to install or keep patched, and a persistent profile.
• captcha solving — see Captchas. None was met in testing; the detection is wired for the day one is.
• fingerprints — a consistent device identity.

---

Install

One engine, in its own virtualenv. The three engines' pins are mutually unsatisfiable (pyee <12 vs >=13, urllib3 <2.0 vs >=2.6), so install exactly one per environment:

git clone https://github.com/2scraper/google-finance-scraper
cd google-finance-scraper
python3 -m venv .venv
./.venv/bin/pip install -r requirements.txt -r requirements-playwright.txt
./.venv/bin/playwright install chromium
./.venv/bin/python smoke_test.py          # offline checks, no network

Use

# one instrument, or several — the unit of work on this site is the SYMBOL
python3 playwright_scraper.py --mode quote \
    --symbols GOOGL:NASDAQ,BMW:ETR,EUR-USD --out quotes

# every strip the market page publishes: indices, sectors, FX, crypto, futures
python3 playwright_scraper.py --mode markets --market US --out markets

# just one strip
python3 playwright_scraper.py --mode markets --market DE --category currency

# the day's movers
python3 playwright_scraper.py --mode movers --market GB --out movers

# quarterly financials, back to 2004 on a large cap
python3 playwright_scraper.py --mode financials --symbols GOOGL:NASDAQ

# the consensus, the target range, and every published analyst action
python3 playwright_scraper.py --mode analysts --symbols GOOGL:NASDAQ

# OHLCV — the latest session at 5-minute resolution AND a month of daily
# bars, in one run with no extra fetch
python3 playwright_scraper.py --mode chart --symbols GOOGL:NASDAQ,BMW:ETR

# who reports this week
python3 playwright_scraper.py --mode earnings --market US

Symbol spelling, and the one trap

kind	form	example
equity, index, futures	TICKER:EXCHANGE	GOOGL:NASDAQ, .INX:INDEXSP, GCW00:COMEX
currency pair, crypto	BASE-QUOTE	EUR-USD, BTC-USD

The colon form of a pair is the trap. EURUSD:CURRENCY does not error — it returns HTTP 200 with Google's own Page Not Found, and the string EURUSD still appears 17 times in those bytes. The scraper reports not_found with the symbol named, rather than an empty result.

---

Modes

Instrument modes — one --symbols entry each, one page fetch each:

mode	reads	rows (GOOGL:NASDAQ, 2026-09-22)
quote	the current quote	1 — open/high/low, volume, market cap, industry, extended hours
financials	one row per reporting period	90 quarters back to 2004 — revenue, net income, operating expense, EBITDA, EPS, net margin, effective tax rate, and what the street estimated
analysts	the consensus and every published action	44 — verdict, buy/hold/sell split, 12-month target range, plus each firm, date, rating and price target
chart	every OHLCV bar the page carries	99 — 79 intraday (5-minute) and 20 daily

Market modes — one page fetch, geo-selected by --market:

mode	reads	rows (gl=US, 2026-09-22)
markets	every strip on the market page	46 — 20 indices, 11 sectors, 5 FX, 5 crypto, 5 futures
movers	that page's gainers / losers / most-active	11
earnings	its upcoming announcements calendar	5, with revenue and EPS estimates

Everything above comes out of the same first response. The site's Overview / Analysis / Earnings / Financials tabs are rendered client-side from data that already arrived, so --mode financials costs exactly one page fetch — the same one --mode quote makes.

Three limitations worth reading before you plan around them

movers returns a PREVIEW, not a ranking. Google retired the standalone /finance/markets/gainers, /losers, /most-active, /currencies, /cryptocurrencies, /climate-leaders and /indexes pages — all seven now redirect to the market page, checked 2026-09-22 — and what that page publishes is the top 1 to 4 of each list, measured across six markets. A full top-50 gainers ranking is not available from this site, at any price, to anyone. This repo reads what Google publishes and says so rather than letting you infer a truncation bug.

The same change is why markets is the valuable mode here: the currency, crypto and futures strips it returns are what those retired URLs used to serve, and they are still published in full.

--window does not deepen the chart. Every value from 5D to MAX returns the same ~20 daily bars, because the deeper history is fetched client-side by an endpoint this repo does not implement. What you get for free is one full session at five-minute resolution plus about a month of daily bars.

financials exposes seven figures, not a full statement. Google publishes 106 unlabelled numbers per quarter; seven of them were identified by matching the page's own rendered rows against every slot across four periods on four instruments in four currencies. The other 99 are left alone, because a mislabelled financial figure is worse than a missing one. EPS is null on instruments where Google publishes none — measured 0 of 88 periods on 7203:TYO, against 90 of 90 on GOOGL:NASDAQ — and is deliberately not filled in from the adjacent basic-EPS slot, which is a different measure.

---

Output

Five row classes, one per kind of thing. quote, markets and movers share Quote (32 columns); financials, analysts, earnings and chart each have their own. All five open with the same five columns — source, scraped_at, url, sku, title — so sku joins every mode to every other, and diff_runs.py refuses to compare two modes.

The quote row, 32 columns, same JSON and CSV field order. See sample_output.json — cut from a real run, not written by hand.

{
  "source": "google.com/finance",
  "url": "https://www.google.com/finance/beta/quote/GOOGL:NASDAQ",
  "sku": "GOOGL:NASDAQ",
  "title": "Alphabet Inc Class A",
  "price": 354.97, "currency": "USD",
  "ticker": "GOOGL", "exchange": "NASDAQ", "instrument_type": "stock",
  "prev_close": 349.54, "open": 350.64, "day_high": 357.61, "day_low": 349.1,
  "volume": 152020, "market_cap": 4312487948430.176,
  "industry": "Interactive media", "country": "US",
  "timezone": "America/New_York",
  "after_hours_price": 356.95,
  "listing": null, "market": null, "price_source": "af_quote"
}

Notes that will save you a wrong conclusion:

• sku is Google's own canonical symbol, read out of the payload rather than rebuilt from the URL — because a URL's symbol is whatever you typed, and you can type one the site does not have.
• market_cap and volume are null, never 0. Google writes 0 where it means "there is none" — on every index, currency, crypto and futures row measured. Written through, that zero drags any average you compute, so both are nulled at the sentinel.
• prev_close is not an original_price. A quote below its previous close is not discounted; there is deliberately no discount_pct column here.
• quoted_at is Google's as-of, scraped_at is ours. A stale quote is visible only in the difference.
• currency is null on an index. An index level is not money, and nothing here defaults to "USD".
• market is load-bearing. The market page's lists are geo-selected, so a gl=US run and a gl=DE run are two SAMPLES, not a before and an after. diff_runs.py refuses to compare across it.

Exit codes

0 ok · 1 crash · 2 bad usage · 3 blocked · 4 zero rows from a page that was fetched · 5 the content was never obtained · 6 partial.

4 and 5 are deliberately different: an empty market and a dead proxy must not be one value to a pipeline.

---

How it reads the page

Google Finance renders its whole payload into AF_initDataCallback({key: 'ds:N', data:[...]}) blobs — plain nested JSON, fully server-rendered, in the first response. There is no JSON-LD at all: measured zero application/ld+json blocks across 11 captures.

The parser never anchors on ds:N, and that is measured. The key numbering is stable across repeated fetches of the same URL and shifts the moment a query parameter changes the page: ?window=5D produces 23 keys instead of 22 and moves every key up by one. Anchoring on a key number is the same mistake as anchoring on a build-hash CSS class, so everything is found by the SHAPE of the record instead. The suite pins this by renumbering every key in a real fixture and asserting no row changes.

Because the payload is in the first response, there is no scroll loop and no hydration wait — an HTTP client with a browser User-Agent and headless Chromium parse to the same rows.

There is no pagination, anywhere

A quote page is one instrument; the market page carries its lists in full. --pages above 1 is refused with the reason rather than honoured, which would fetch one page three times and report a complete three-page run. --concurrency splits symbols, not pages.

---

Blocks, and what actually gets refused

The thing Google refuses here is usually the client, not the address — and the direction is the opposite of what this family's other sites taught. Same address, same URL, one fetch each, 2026-09-22:

client	result
curl, curl's own User-Agent	redirected to /finance/beta/unsupported, 657 KB, no data
curl, a Chrome User-Agent	the real page, 1.31 MB
headless Chromium	the real page

Sec-Fetch-Dest/Mode/Site made no difference either way. So on this site a browser User-Agent is required, where a sibling site refuses curl precisely for wearing one. Every engine here sets one, so if you ever see unsupported_client in the sidecar, something overrode it — that state is reported as its own thing and never as a block, because no proxy, exit country or captcha solve would change it.

A real block is Google's /sorry/index interstitial, which is about the address. None was met in 11 fetches from a datacentre address.

Captchas

No captcha was rendered in testing. Every candidate marker scored zero across all 11 captures, served and refused alike.

That is a measurement from one address on one day, and explicitly not a claim that Google does not challenge scrapers. It serves /sorry/index with a reCAPTCHA to addresses it scores badly; this repo recognises that page, and 2Captcha solves that task type (RecaptchaV2TaskProxyless / RecaptchaV3TaskProxyless). The solver stays wired and costs nothing while no challenge appears — --solve-captcha when-blocked is the default and counts instrument links on the spot rather than paying for a page whose rows are already there.

---

Engines

All three produce identical rows: verified 2026-09-22 on the same URL, 46 rows each, zero differing stable columns between Playwright and its twins.

engine	notes
Playwright	primary. The only one with --concurrency.
Selenium	cannot use an authenticated remote CDP endpoint (debuggerAddress takes a bare host:port), and --proxy-server cannot authenticate at all — credentials are stripped with a warning.
Puppeteer (pyppeteer)	effectively unmaintained upstream; its own README points at Playwright. Adds --chromium-path.

Configuration

Credentials go in .env beside the scripts, never on a command line — anything that can run ps reads an argv, and shell history keeps it.

cp .env.example .env
python3 env_config.py     # prints what was picked up, without printing secrets

Precedence, highest first: explicit flag → exported environment variable → .env → default. A copied .env.example reads as unset: every placeholder still containing {...} is treated as absent, so it is never sent to an API as if it were a key.

---

Development

python3 smoke_test.py            # 268 offline checks, no network
python3 .github/ci_checks.py --all
python3 make_fixtures.py         # regenerate fixtures from your own captures

Fixtures live in fixtures_generated.json, cut from real captures by make_fixtures.py, which proves each trimmed fixture parses identically to the capture it came from — column for column — before writing it.

captures/ is deliberately not in the repository: eleven captures of this site are 13 MB.

Licence

MIT. Not affiliated with, endorsed by, or connected to Google. Respect the site's terms and the law where you operate; this is a tool, and what you do with it is yours.