Scrape Glovo Spain: every store of the restaurant chains you choose in Madrid or Barcelona, with delivery fee, service fee, rating, promotions and the price of each menu product per store. No login.
First version. Glovo stores of a chain per city found with an adaptive grid of delivery points,
and the menu of each store. Plain HTTP to api.glovoapp.com, no login, no browser, no proxy.
Store records (one per storeId + addressId) and product records (one per addressId +
storeProductId).
Source behaviour, as measured on 2026-11-15 from a home connection:
GET /v3/stores/{slug} with the delivery point in headers: 200 with the store that serves the
point, 404 when none does.
GET /v4/stores/{storeId}/addresses/{addressId}/content/main: ~400 KB, ~100 KB on the wire.
A product listed in two sections ("Promociones" and its own) appears twice: one record, both
sections.
POST /v1/web/store_wall/search: with the headers listed in the survey it answered
503 no available server
. The web client's own header set (read from the page on 2026-11-15: request
id, perseus client id and consent, dynamic session id, device URN, app context, request TTL,
and Glovo-Client-Info: web-customer-web-react/{version} project:customer-web) gives 200. All
ids are random UUIDs per run; no cookie, token or key.
App version v1.2968.0, read from NEXT_PUBLIC_APP_VERSION in the city page on every run.
City codes: MAD and BCN verified. VLC and SVQ answer 400; the real codes were not found,
so Valencia and Sevilla are not offered.
Menu throttling: at one menu every 0.5 s, 429 (code: throttle, domain store-view) after
about 25 menus; 14 throttled answers and 2 menus lost out of 83. At 1.5 s: 83 of 83, no 429. Menus
now have their own pacing (menuIntervalSecs, default 1.5).
Live runs (tools/live_check.py, home connection):
Run
Points
Served
Stores
Products
Time
Errors
Burger King Madrid, grid 2 km + 0.5 km
2,308 (196 coarse)
1,392
83
—
1,154 s
0, no 429
Burger King Madrid, 83 menus at 1.5 s
—
—
—
16,941
123 s, 8.5 MB
0, no 429
Goiko Madrid, same grid
1,988
1,016
20
—
994 s
0, no 429
Goiko Madrid, 20 menus
—
—
—
1,355
29 s, 0.8 MB
0
Burger King Madrid, second sweep (~75 min after the first)
2,308
1,392
79
—
1,154 s
0, no 429
Id stability: the 79 stores of the second sweep all have the same storeId:addressId as
in the first, and 71 of them the same first grid point. The store set depends on the time of
day: the 4 stores missing in the second sweep (00:05 vs ~01:20) were open in the first; when a
store closes Glovo assigns its points to another one. Schedule sweeps at a fixed hour inside
opening hours (e.g. 13:00) so runs are comparable.
Goiko: 20 stores, 31 of 70 products with more than one price across stores (the survey
found 31 of 66 between two stores).
Coverage: the survey's full 0.5 km grid (3,080 points) found 77 Burger King stores; the
adaptive grid found 83 with 2,308 points (75 % of the calls). Burger King delivers almost
everywhere in Madrid, so few coarse cells are skipped; chains with fewer stores save more.
Price per store: 119 of 227 Burger King products (by SKU) have more than one price across
the 83 stores.
75 of the 83 stores have a 280xx postal code; the rest are just outside the municipality but
deliver into it.
Tests: 56 offline checks (tests/test_v01.py): headers, store and menu parsing, search tiles,
grid geometry, slug table, privacy (no phone; fiscal name only for companies), dataset schema,
and a stubbed run with a 404-everywhere slug, a 503 slug and a failing menu.
Not in 0.1:
matching Glovo stores to locator stores by address (Glovo gives no coordinates);
Valencia, Sevilla and other cities (city codes not found);
datacenter-IP test of this header set (first platform run).
Platform smoke test (2026-11-15, Apify datacenter IP, no proxy, build 0.1.1 beta): inconclusive.
Run UXia0xEniDo3Taxpu (08:06 UTC), Goiko Madrid capped at 20 points: the city page answered
200 (app version v1.2968.0 read) and the 20 store requests answered 404, not 403. No block
seen, but no store either: with the cap, the 20 points are the first rows of the grid, on the
south edge of the box, where Goiko does not deliver. The run failed as designed
("404 at every point"). The smoke input with maxPointsPerSlug: 20 was a bad choice.
Run R9f751TWgM3NI8Uam (Burger King, full coarse grid) was aborted at the owner's request about
a minute after it started. Datacenter access to stores/{slug} and the menus is still unproven.
Run ZjeDFXQleg73tfid9 (08:13 UTC), Apify RESIDENTIAL ES proxy, Burger King, coarse grid
4 km, 1 menu: the city page answered through the proxy (app version read in 1.5 s), then no
store request finished: no warning in 5 minutes and the run hit the 300 s platform timeout
(TIMED-OUT, no SUMMARY, USD 0.03). With a 60 s socket timeout and 4 retries, one request that
never answers takes about 330 s, so the very first api.glovoapp.com call hung through the
proxy. Corrected later the same day: the actor did not log successful points, so a slow
proxy (several seconds per point) looks the same as a hang. This run proves neither.
Run 8356m0ogh2MarMLEe (08:21 UTC), Apify datacenter, no proxy, Burger King, coarse grid
2 km: app version read, then the first ~64 s of store requests (about 120 at one every 0.5 s)
answered without error; from 08:22:50 every request got 403 with a CloudFront-style HTML
"ERROR" page (195 in a row, none recovered). Aborted at 08:24. Reading: a volume/rate block on
the datacenter IP after about a minute, not a header problem (the home run did 2,308 requests at
the same pace with no error). From datacenter the sweep does not work as is.
"Required request header 'Glovo-Perseus-Session-Id' is missing" is what the API answers to a
request without the web headers (e.g. opening the URL in a browser tab); the actor sends that
header on every API call.
Residential proxy (Decodo, ASN 12338, Spain) as a secret env var of the actor:
From home, through Decodo (08:33 UTC): city page 3.3 s, 3 store requests 200 in 0.8-1.2 s.
Run wbDnfe6pxbt18VcJl (08:29 UTC), coarse grid 2 km (196 points), 240 s timeout:
TIMED-OUT with no warning. At about 1.2 s per point the grid needs about 4 minutes: the timeout
was too short for the run, not a block.
Run SYBgg86Qbw2Kqdu1n (08:34 UTC), Burger King, coarse grid 4 km, 1 menu: SUCCEEDED.
49 points, 21 served, 16 stores, 0 errors, 1 network error retried; 75 s, 0.1 MB.
The menu answered 200 with "no product list for this store": at 09:35 Madrid time the store
was closed. Menus through the proxy still have to be checked inside opening hours.
Code changes after the platform tests (still 0.1, builds 0.1.2-0.1.4 on beta):
Proxy from the secret env var GLOVO_PROXY_URL (URL or Decodo host:port:user:pass), used when
the input asks for no proxy. The URL is never logged; SUMMARY.egress says direct,
env-proxy or input-proxy.
fineStepKm: 0 now means coarse grid only, as the schema says. It used to fall back to 0.5.
Progress log every 25 points. Without it a slow run looked like a hung one.
SUMMARY.menus lists each menu with its status and error; menusEmpty counts closed stores.
Egress fallback (egressMode, default auto): direct first; on the first 403 the client
switches once to the residential proxy of GLOVO_PROXY_URL and resends that request; no
rotation, no other change to the request. egressMode: proxy uses the proxy from the start. An
input proxyConfiguration overrides both. Tests: 67 offline checks (switch on 403, no second
switch, 403 without fallback fails).
Run QwB2Q6BJnmhqrHabg (2026-11-15 08:45 UTC, build 0.1.5 beta, egressMode: auto), Burger
King, coarse grid 2 km, 1 menu: SUCCEEDED. Direct (Apify datacenter) for the first 135
requests (~67 s), then one 403; the client switched to Decodo and the rest went through it with
0 errors. 196 points, 82 served, 43 stores; 1 menu ok with 222 products through the
proxy; 168 s, 0.3 MB. Second datacenter block measured at the same point (~120-135 requests,
~1 minute) as run 8356m0ogh2MarMLEe.