Scrape App Store and Google Play reviews in Python with typed sync and async clients for review ingestion, app search, metadata lookup, and resumable cursors. Use public sources without credentials, or supply store credentials for the official review APIs.
Documentation · Comparison · FAQ · Changelog
from app_reviews import AppStoreReviews, Country
with AppStoreReviews() as client:
result = client.fetch(
"324684580", # numeric Apple app ID from its App Store URL
countries=[Country.US, Country.GB],
limit=20,
max_pages=2,
)
for review in result:
print(review.country, review.rating, review.body[:80])
if result.errors:
print(result.errors)- One typed model for Apple App Store and Google Play reviews.
- Sync and native async methods for fetch, paging, search, and lookup.
- Resumable cursors, streaming iterators, request budgets, retry policy, and connection pooling.
- Partial failures remain visible through typed errors and per-source outcomes.
- A complete JSON-safe result envelope for automation.
- Search and metadata lookup for both stores.
- Developer replies through both official APIs, with safe single-attempt writes.
- App Store version history and App Store Connect versions with release notes.
- Python 3.11+ with
py.typed; runtime dependencies arehttpxandcryptography.
The public endpoints are unofficial and can be throttled or changed by the stores. Authenticated sources require an Apple Developer or Google Play Developer account. See source capabilities before choosing a source.
pip install app-reviewsOr:
uv add app-reviewsThe credential-free client uses Apple's public RSS feed. Reviews are partitioned
by storefront, so countries= controls which storefronts are fetched.
from app_reviews import AppStoreReviews, Country
with AppStoreReviews() as client:
result = client.fetch(
"324684580",
countries=[Country.US, Country.DE],
limit=100,
max_pages=5,
)Use the numeric Apple app ID—the number after /id in an App Store URL. Search
and lookup also accept the identifiers documented in the
Python API guide.
Google Play reviews form one global corpus. A review response does not include a
reviewer country, so Review.country is None; a store presentation country is
not a review filter.
An explicit empty or all-blank countries collection is a no-op and makes no
requests on any review client. Otherwise, Google Play review clients reject any
nonblank country or countries selection before network I/O because Play
reviews are global. Google Play search and metadata still accept country to
select the storefront used for availability, presentation, and price.
from app_reviews import GooglePlayReviews
with GooglePlayReviews() as client:
result = client.fetch(
"com.spotify.music",
limit=100,
max_pages=5,
)Use the package name from the id query parameter in a Google Play URL.
| Client configuration | Source | Review country behavior | Credentials |
|---|---|---|---|
AppStoreReviews() |
Apple RSS | One corpus per requested storefront | No |
AppStoreReviews(auth=...) |
App Store Connect | Global request; territory may be present on each review | Yes |
GooglePlayReviews() |
Google Play web | Global; country=None |
No |
GooglePlayReviews(auth=...) |
Google Play Developer API | Global; country=None |
Yes |
Review.language is populated only when a provider reports a language. The
Google Play Developer API can report reviewerLanguage; do not interpret that
as reviewer location. Google Play web reviews have no title, while the Developer
API may expose a legacy title embedded in its review text.
Search and lookup are credential-free. Here, country selects the storefront
used for availability, presentation, and price; it still does not identify a
reviewer's country.
from app_reviews import AppStoreSearch, Country, GooglePlaySearch
with AppStoreSearch() as apple:
ios_apps = apple.search("fitness tracker", country=Country.US, limit=5)
with GooglePlaySearch() as play:
android_apps = play.search("fitness tracker", country=Country.US, limit=5)
print([app.name for app in ios_apps + android_apps])Search returns list[AppMetadata]; lookup returns AppMetadata | None.
AppMetadata.release_notes carries the current version's "What's New" text: on
every App Store result, and from Google Play lookup().
The iTunes APIs report only the current version. version_history() reads the
"Version History" from the public App Store product page instead, newest first:
from app_reviews import AppStoreSearch
with AppStoreSearch() as apple:
for entry in apple.version_history("324684580"):
print(entry.version, entry.released_at.isoformat(), entry.release_notes)Each AppVersionEntry has version, released_at (timezone-aware UTC), and
release_notes. It takes the numeric app ID and goes through the client's
proxy=, retry=, and rate_limiter= settings. An app the store does not
have, or a page with no history, returns []; a history that cannot be read
raises ParseError. aversion_history() is the async twin.
Scraped source.
version_history()reads the public product page, not an API. It is best-effort and App Store only, and it may break when Apple changes the page.AppStoreVersionsreads versions from the official App Store Connect API, but that API has no release date, so this stays the source for dates.
fetch() returns FetchResult, which is iterable over Review objects. It also
records one CountryOutcome for each source walk. A fetch may contain reviews
and errors at the same time.
for outcome in result.outcomes:
print(
outcome.country,
outcome.pages,
outcome.reviews_fetched,
outcome.skipped_reviews,
outcome.stopped_because,
)
for error in result.errors:
if error.retryable:
schedule_retry(error)stopped_because="exhausted" means the source reported no next page. Reasons
such as limit, since, and max_pages mean the client stopped while more data
may exist. Malformed records that can be isolated are counted as
skipped_reviews; malformed page envelopes are parse errors.
Apple's RSS JSON feed sometimes answers with no entries while its XML feed for
the same page has them. The package then reads the XML feed, through the same
client and settings, and records it: outcome.feed_format is "xml" (else
"json", and None for other sources). Both empty is a normal "exhausted".
fetch() retains partial failures as data. Search and lookup raise typed
exceptions such as RateLimitError, RequestError, NotFoundError, and
ParseError instead of returning fetch diagnostics.
Use result.to_dict() when diagnostics matter. It returns the full JSON-safe
envelope: reviews, outcomes, errors, and skipped_reviews. Use
result.to_dicts() only when review rows are all you need. Provider raw
payloads are omitted unless include_raw=True.
import json
payload = result.to_dict()
print(json.dumps(payload, indent=2))Guard an empty result before deriving CSV headers:
import csv
rows = result.to_dicts()
if rows:
with open("reviews.csv", "w", newline="", encoding="utf-8") as stream:
writer = csv.DictWriter(stream, fieldnames=list(rows[0].keys()))
writer.writeheader()
writer.writerows(rows)Agents that can call Python functions can use the library directly; an extra package interface is not required. Keep the function bounded and return the complete envelope so the caller can distinguish an empty success, a partial result, and a provider failure.
from typing import Any
from app_reviews import AppStoreReviews, Country
def fetch_app_store_reviews(app_id: str, country: str = "us") -> dict[str, Any]:
"""Fetch at most 50 reviews across at most two provider pages."""
with AppStoreReviews() as client:
result = client.fetch(
app_id,
countries=[Country(country.lower())],
limit=50,
max_pages=2,
)
return result.to_dict()Register that function with the agent framework used by your application. Keep authentication, allowlists, timeouts, retry decisions, and user authorization in the application layer. The package itself is a Python library and does not run a network service or accept prompts.
Every network entry point has an async twin. Use iter_reviews() /
aiter_reviews() to avoid buffering a multi-storefront fetch, or iter_pages() /
aiter_pages() when you need to persist next_cursor.
import asyncio
from app_reviews import GooglePlayReviews
async def main() -> None:
async with GooglePlayReviews() as client:
result = await client.afetch(
"com.spotify.music",
limit=100,
max_pages=3,
)
print(len(result), result.errors)
asyncio.run(main())For Apple RSS, multi-country fetch() caps default fan-out at eight workers.
Pass concurrency= to choose a smaller or larger explicit limit.
concurrency= limits parallel country walks within one fetch. When one process
fetches many apps, pass the same RateLimiter to every client instead, so they
share one request budget. It is thread-safe and asyncio-safe, and every attempt,
retries included, takes a token.
from concurrent.futures import ThreadPoolExecutor
from app_reviews import AppStoreReviews, RateLimiter
limiter = RateLimiter(rate=2.0, burst=4) # two requests per second, bursts of four
def fetch(app_id: str):
with AppStoreReviews(rate_limiter=limiter) as client:
result = client.fetch(app_id, countries=["us", "gb", "de"], max_pages=2)
throttled = [error.country for error in result.errors if error.kind == "rate_limited"]
return result, throttled
with ThreadPoolExecutor(max_workers=16) as pool:
results = list(pool.map(fetch, ["324684580", "310633997"]))When a store throttles (HTTP 429, or the HTTP 403 Apple's RSS feed answers while
it blocks an address), the limiter pauses every client sharing it: for the
server's Retry-After if it sent one, else 30 seconds, doubling on each
consecutive throttled answer up to 15 minutes. A successful answer resets the
doubling. Tune it with RateLimiter(rate, burst, initial_penalty=30.0, max_penalty=900.0), or pause it yourself with limiter.penalize(seconds).
Any object with acquire(), aacquire(), and record(status, retry_after)
satisfies the RequestLimiter protocol and can be passed instead, for a budget
shared across processes. record runs once per response.
Throttled storefronts fail alone: other countries keep their reviews, and each
throttled country carries FetchError(kind="rate_limited", retryable=True) in
its CountryOutcome. The package does not retry those 403s itself, because
requests made during a block extend it; fetch the throttled countries again
later through the same limiter.
Pass AppStoreAuth to use App Store Connect or GooglePlayAuth to use the
Google Play Developer API. The authenticated APIs only expose apps owned by the
credential holder and have different history, ordering, and field behavior.
from app_reviews import AppStoreAuth, AppStoreReviews
auth = AppStoreAuth(
key_id="ABC123DEF4",
issuer_id="12345678-1234-1234-1234-123456789012",
key_path="/path/to/AuthKey.p8",
)
with AppStoreReviews(auth=auth) as client:
result = client.fetch("324684580", limit=100, max_pages=3)The key can also come from memory, for a secret manager or an environment
variable: AppStoreAuth(key_id=..., issuer_id=..., private_key=pem_text) and
GooglePlayAuth(service_account_info=parsed_json). Each takes exactly one of its
two sources, and neither shows the key in its repr.
See authentication for Google credentials and setup details.
AppStoreReplies and GooglePlayReplies write the public developer reply to a
review. Both need credentials that are allowed to answer reviews:
| Store | Credential | Permission |
|---|---|---|
| App Store | App Store Connect API key | a team key with the Customer Support or Admin role (Apple) |
| Google Play | Service account invited in Play Console | Reply to reviews for the app |
from app_reviews import AppStoreReplies, GooglePlayAuth, GooglePlayReplies
with AppStoreReplies(auth) as apple:
if apple.get_reply(review.id) is None:
reply = apple.reply(review.id, "Thanks, fixed in 4.2.")
print(reply.state) # "pending": Apple can take up to 24 hours to show it
play_auth = GooglePlayAuth(service_account_path="/path/to/service-account.json")
with GooglePlayReplies(play_auth) as play:
play.reply("gp:AOqpTO...", "Thanks for the report!", package_name="com.example.app")reply() creates the reply or replaces the existing one and returns a
ReviewReply(review_id, reply_id, text, state, updated_at). state is
"pending" until Apple shows the reply and "published" after; Play replies are
published at once and have no reply_id. get_reply() returns the current
ReviewReply or None. AppStoreReplies.delete_reply() removes a reply and
returns whether there was one; Play has no delete API. Each method has an async
twin: areply(), aget_reply(), and adelete_reply().
Each reply write is attempted once, whatever retry= says, to avoid blindly
repeating a write that may already have succeeded. Check get_reply() before
replying again after any failure:
ReplyOutcomeUnknownError: a timeout, dropped connection, 5xx, or redirect after the request was sent (writes never follow redirects, which would re-send them). The reply may or may not be live.ReplyRejectedError(reason): the store refused it, and nothing was published. A Play reply over 350 characters raisesreason="too_long"before anything is sent.RateLimitError: HTTP 429;retry_afteris the wait the store asked for.AuthError: the key is unusable or lacks the permission.
Reads (get_reply()) follow the normal retry policy.
AppStoreVersions(auth).versions(app_id) lists an owned app's versions from the
official API, newest created_at first, each with its "What's New" text per
locale:
from app_reviews import AppStoreVersions
with AppStoreVersions(auth) as client:
for version in client.versions("6741066976"):
print(version.version, version.state, version.release_notes.get("en-US"))The official API has no release date. AppStoreVersion.created_at is when the
version was created in App Store Connect, and earliest_release_date is only the
earliest moment a scheduled release may go out. For the dates versions reached
the store, AppStoreSearch.version_history() (scraped from the product page)
remains the source. state is Apple's appVersionState, which tells whether a
version is live: "READY_FOR_DISTRIBUTION" once it is on the store.
- Apple RSS exposes roughly 500 recent reviews per storefront and an empty feed cannot distinguish no reviews, an unknown app, and some upstream throttling.
- The Google Play Developer API returns reviews created or modified within the last seven days; this package does not read Play Console CSV exports.
- Google Play's public web endpoint is undocumented and can change without notice.
- Store terms and applicable law depend on how and where you use the data. Review them for your use case.
Read source capabilities, the comparison, and the FAQ for details.
git clone https://github.com/0xfirattamur/app-reviews.git
cd app-reviews
make install
make allSee CONTRIBUTING.md and SECURITY.md.
The Google Play web parsing is based on field-path knowledge from google-play-scraper, reworked on this project's HTTP, retry, paging, and model layers.