Scavio Python SDK

The official Python SDK for the Scavio Search API. Access real-time data from 36 sources -- Google, Amazon, Walmart, Costco, Mercado Libre, eBay, Target, Home Depot, YouTube, Reddit, X, TikTok, TikTok Shop, Instagram, LinkedIn, Threads, Kuaishou, Zillow, Redfin, Realtor.com, Rightmove, Booking.com, Tripadvisor, Airbnb, Yelp, Indeed, Glassdoor, the Apple App Store, Google Play, SEC EDGAR, Companies House, G2, Capterra, Trustpilot, Google Ads Transparency and the Meta Ad Library -- plus extract() for reading any URL, all with a single API key. Built for AI agents, LLM applications, and data pipelines.

One API key, 36 data sources, 233 endpoints, structured JSON with knowledge graphs. A powerful alternative to Tavily, SerpAPI, and ScraperAPI for developers who need more than just web search.

Why Scavio

Tavily focuses on AI-optimized web search. SerpAPI offers SERP parsing across search engines with separate plans. ScraperAPI provides raw web scraping with proxy rotation. Scavio combines multi-source structured data in a single search API for AI agents with one SDK and one API key.

Installation

pip install scavio

Quick Start

Get your free API key at dashboard.scavio.dev.

from scavio import ScavioClient

client = ScavioClient(api_key="sk_...")  # or set SCAVIO_API_KEY env var

results = client.search("best noise cancelling headphones 2026")
for r in results["organic_results"]:
    print(r["title"], r["link"])

Every method returns the API response as a plain dict. Amazon responses are normalized to a stable, documented shape; the other endpoints keep each platform's own field names, so fields vary by endpoint.

Fully typed parameters

Every endpoint exposes all of its parameters as explicit, documented, autocomplete-friendly keyword arguments with Literal types for enums. Your editor shows the full parameter set, allowed enum values, and defaults inline.

# Google web search with the full parameter surface
results = client.google.search(
    "electric cars",
    gl="us",                 # country of the search
    hl="en",                 # UI language
    location="Austin, Texas, United States",
    time_period="last_month",
    device="mobile",
)

# YouTube filters. The digit-named API fields (4k, 360, 3d) are exposed as
# valid Python identifiers: four_k, video_360, video_3d.
client.youtube.search("drone footage", four_k=True, hdr=True, duration="long")

# Amazon product lookup: pass the ASIN (sent to the API as `query`).
# `country` is the marketplace, as an ISO 3166-1 alpha-2 code.
client.amazon.product("B09XS7JWHH", country="gb")

Forward-compatible passthrough

Any parameter the API adds in the future can be passed via **extra and is sent verbatim, so you never have to wait for an SDK release:

client.google.search("openai", **{"some_new_param": "value"})

Retries and resilience

The client automatically retries transient failures (HTTP 429 and 5xx, plus network/timeout errors) with exponential backoff, jitter, and Retry-After support. Configure or disable it with max_retries.

1. AI Web Research -- Feed Search Results to an LLM

from scavio import ScavioClient

client = ScavioClient()

results = client.search("latest advances in quantum computing 2026")

context = "\n\n".join(
    f"[{r['title']}]({r['link']})\n{r.get('snippet', '')}"
    for r in results["organic_results"]
)

prompt = f"Based on these search results, summarize the latest advances:\n\n{context}"
# Pass `prompt` to your LLM of choice (OpenAI, Anthropic, etc.)
print(prompt[:500])

2. Price Comparison -- Amazon vs Walmart

from scavio import ScavioClient

client = ScavioClient()

query = "sony wh-1000xm5"
amazon = client.amazon.search(query, country="us")
walmart = client.walmart.search(query)

print("Amazon:")
for p in amazon["data"]["products"][:3]:
    print(f"  ${p['price']} - {p['title'][:60]}")

print("\nWalmart:")
for p in walmart["data"]["products"][:3]:
    print(f"  ${p['price']} - {p['title'][:60]}")

3. Product Lookup by ASIN, plus every seller offer

from scavio import ScavioClient

client = ScavioClient()

data = client.amazon.product("B0BS1PRC4L")["data"]

print(f"Brand:   {data['brand']}")
print(f"Title:   {data['title']}")
print(f"Rating:  {data['rating']} ({data['reviews_count']} reviews)")
print(f"Price:   {data['price']} {data['currency']}")

# Same ASIN, every seller: price, condition, and who holds the buy box.
offers = client.amazon.offers("B0BS1PRC4L")["data"]
print(f"{offers['total_offers']} offers")
for o in offers["offers"][:5]:
    tag = " (buy box)" if o["is_buy_box_winner"] else ""
    print(f"  {o['price']} {o['currency']} - {o['seller_name']} [{o['condition']}]{tag}")

4. SEO Competitor Analysis

from scavio import ScavioClient

client = ScavioClient()

results = client.search("best project management software", gl="us")

for r in results["organic_results"]:
    print(f"{r['position']}. {r['title']}")
    print(f"   {r['link']}")

5. News Aggregation

from scavio import ScavioClient

client = ScavioClient()

news = client.google.news("AI startups")

for article in news["news_results"][:5]:
    print(f"[{article['source']}] {article['title']}")
    print(f"  {article['link']}")
    print()

6. YouTube Content Discovery

from scavio import ScavioClient

client = ScavioClient()

videos = client.youtube.search("python tutorial", sort_by="view_count")

for v in videos["data"]["results"][:5]:
    print(f"{v['title']} ({v['view_count']:,} views)")
    print(f"  {v['url']}")

# Full details for a specific video (metadata() is a deprecated alias of video())
video = client.youtube.video("dQw4w9WgXcQ")
print(f"\n{video['data']['title']}")
print(f"  {video['data']['view_count']:,} views")

# Transcript, related videos, comments, channel, and streams
transcript = client.youtube.transcript("dQw4w9WgXcQ", format="text")
related = client.youtube.related("dQw4w9WgXcQ")
comments = client.youtube.comments("dQw4w9WgXcQ")
channel_id = client.youtube.channel_resolve("@mkbhd")["data"]["channel_id"]
channel = client.youtube.channel(channel_id)
streams = client.youtube.streams("dQw4w9WgXcQ")

7. Reddit Market Research

from scavio import ScavioClient

client = ScavioClient()

posts = client.reddit.search("best mechanical keyboard")

for post in posts["data"]["results"]:
    print(f"r/{post['subreddit']} - {post['title']}")
    print(f"  {post['url']}")
    print()

# Drill into a subreddit, a single post, or a redditor. reddit.post() takes a
# url or a post_id and returns the post alone -- comments are a separate call.
feed = client.reddit.subreddit_posts("MechanicalKeyboards", sort="TOP")
detail = client.reddit.post(post_id="t3_1v6ngaf")
comments = client.reddit.post_comments("t3_1v6ngaf", sort="TOP")
history = client.reddit.user_posts("spez")
popular = client.reddit.popular()
trending = client.reddit.trending()

8. TikTok Hashtag Analysis

from scavio import ScavioClient

client = ScavioClient()

hashtag = client.tiktok.hashtag(hashtag_name="python")
info = hashtag["data"]["challengeInfo"]

print(f"#{info['challenge']['title']}")
print(f"  Views: {int(info['statsV2']['viewCount']):,}")
print(f"  Videos: {int(info['statsV2']['videoCount']):,}")

9. Instagram Profile and Posts

from scavio import ScavioClient

client = ScavioClient()

profile = client.instagram.profile(username="instagram")
user = profile["data"]["user"]
print(f"@{user['username']} - {user['edge_followed_by']['count']:,} followers")

posts = client.instagram.user_posts(username="instagram", count=12)
reels = client.instagram.user_reels(username="instagram")
hashtags = client.instagram.search_hashtags("fashion")

10. X Search and Profiles

from scavio import ScavioClient

client = ScavioClient()

tweets = client.x.search("AI agents", search_type="Latest")
for t in tweets["data"]["timeline"][:5]:
    print(f"@{t['screen_name']}: {t['text'][:80]}")

# Profile, a user's tweets, followers, and a single tweet's replies
profile = client.x.user("elonmusk")
timeline = client.x.user_tweets("elonmusk")
followers = client.x.user_followers("elonmusk")
replies = client.x.tweet_comments("1808168603721650364", rank="top")
trending = client.x.trending(country="UnitedStates")

11. LinkedIn People and Companies

from scavio import ScavioClient

client = ScavioClient()

# Member profile (1 credit) and their recent posts (10 credits per page). A
# handle or a full LinkedIn URL works anywhere.
person = client.linkedin.person(username="williamhgates")
person_posts = client.linkedin.person_posts(url="https://www.linkedin.com/in/williamhgates/")

# Company profile and its recent posts
company = client.linkedin.company(company="microsoft")
company_posts = client.linkedin.company_posts(company="microsoft")

# Jobs: search, then pull full detail for one listing
job_results = client.linkedin.search_jobs("software engineer", location="United States")
job = client.linkedin.job(job_id=job_results["data"]["data"][0]["id"])

# A post and its comments (10 per page)
post = client.linkedin.post(post_id="7488618410256523265")
comments = client.linkedin.post_comments(post_id="7488618410256523265", page=1)

Retired endpoints. The LinkedIn data behind person_contact, company_people, company_jobs, search_people and search_posts is no longer available. They remain callable but always return HTTP 410 and are never billed. company() still returns featured_employees (a small sample of staff), and search_jobs() with a company name substitutes for company_jobs.

12. TikTok Shop Product Research

from scavio import ScavioClient

client = ScavioClient()

# Listings carry exact prices
results = client.tiktok_shop.search("phone case")
for p in results["data"]["products"][:5]:
    print(p["title"], p["price"]["current"], p["shop"]["shop_name"])

# Detail adds description, variants, stock and shipping -- but NOT a price
# (TikTok masks it), and it resolves only about 44% of the ids search returns.
# A 404 there is a normal outcome, not an error: skip the item, do not retry.
from scavio import NotFoundError

product_id = results["data"]["products"][0]["product_id"]
try:
    detail = client.tiktok_shop.product(product_id)
    print(detail["data"]["title"], len(detail["data"]["variants"]), "variants")
except NotFoundError:
    pass  # TikTok Shop has no detail data for this product; skip it, do not retry

reviews = client.tiktok_shop.product_reviews(product_id, page_size=200, sort="relevant")
catalog = client.tiktok_shop.shop_products("7495514739648989419")   # exact prices
tree = client.tiktok_shop.categories()
resolved = client.tiktok_shop.resolve("https://vt.tiktok.com/ZT2AHoGsE/")

13. Social Media Monitoring

from scavio import ScavioClient

client = ScavioClient()

brand = "scavio"
reddit = client.reddit.search(brand)
tiktok = client.tiktok.search_videos(brand, count=5)

print(f"Reddit mentions ({len(reddit['data']['results'])}):")
for post in reddit["data"]["results"][:3]:
    print(f"  r/{post['subreddit']}: {post['title']}")

tiktok_videos = tiktok["data"].get("search_item_list", [])
print(f"\nTikTok mentions ({len(tiktok_videos)}):")
for v in tiktok_videos[:3]:
    desc = v["aweme_info"].get("desc", "No description")
    print(f"  {desc[:80]}")

14. Price Drop Alert

from scavio import ScavioClient

client = ScavioClient()

product = client.walmart.product("123456789")
price = product["data"]["price"]
title = product["data"]["title"]

threshold = 50.00
if price and price < threshold:
    print(f"PRICE DROP: {title[:60]}")
    print(f"  Now ${price} (threshold: ${threshold})")
else:
    print(f"{title[:60]}: ${price}")

15. Async Multi-Source Search

import asyncio
from scavio import AsyncScavioClient

async def main():
    async with AsyncScavioClient() as client:
        google = await client.search("mechanical keyboard")
        amazon = await client.amazon.search("mechanical keyboard", country="us")

        print(f"Google: {len(google['organic_results'])} results")
        print(f"Amazon: {len(amazon['data']['products'])} products")

        for r in google["organic_results"][:3]:
            print(f"  Web: {r['title'][:60]}")
        for p in amazon["data"]["products"][:3]:
            print(f"  Amazon: ${p['price']} - {p['title'][:50]}")

asyncio.run(main())

16. Check API Usage

from scavio import ScavioClient

client = ScavioClient()

usage = client.get_usage()
print(f"Plan: {usage['plan']}")
print(f"Credits remaining: {usage['credit_balance']}")

17. Read Any URL -- extract()

extract is a core capability, not a platform, so it is a top-level method: client.extract(url), never client.extract.extract(). It is the "read this page" primitive an agent reaches for when the page is not on a site Scavio has a dedicated namespace for.

from scavio import ScavioClient

client = ScavioClient()

# Default: readability Markdown, normal mode. 1 credit.
page = client.extract("https://example.com/blog/post")["data"]
print(page["content_length"], "chars of", page["format"])
print(page["content"][:500])

# format="html" is the raw page; format="text" is the Markdown flattened.
raw = client.extract("https://example.com/blog/post", format="html")

# mode is the price-bearing parameter, and the only knob that matters on a
# hard target: "normal" (1 credit) -> "advanced", a full browser render
# (1 credit) -> "ultra", the hardest-target tier (2 credits).
hard = client.extract("https://example.com/spa", format="markdown", mode="ultra")

# Billing is charged only on a successful extraction: a dead link, a bot wall
# or a timeout costs nothing, so retrying up a tier is safe.

18. eBay Price Research -- Sold Listings, Not Asking Prices

Live listings tell you what sellers want. sold=True searches completed listings that actually sold, which is what a pricing model needs.

from scavio import ScavioClient

client = ScavioClient()

sold = client.ebay.search(
    query="airpods pro 2",
    sold=True,
    condition="used",
    per_page=240,          # 60, 120 or 240 only; anything else falls back to 60
    min_price=50,          # prices are numbers, so 49.99 is legal too
)["data"]

prices = sorted(p["price"] for p in sold["products"] if p.get("price"))
print("median sold price:", prices[len(prices) // 2])

# On the sold view eBay publishes no headline count, so total_results is null.
# Page with `page` until a page comes back empty rather than trusting a total.
assert sold["total_results"] is None

# On the LIVE view total_results is eBay's own estimate and an unstable one --
# four identical requests minutes apart reported 28k, 140k, 170k and 27k.
# Treat it as an order of magnitude, never as a count.

# `seller` works with no query at all, which is how you page a whole catalogue.
# ebay.seller() is a profile card only -- it cannot enumerate inventory.
catalogue = client.ebay.search(seller="mytechstore", page=1)["data"]
profile = client.ebay.seller("mytechstore")["data"]

19. Meta Ad Library -- Walking a Competitor's Whole Ad Set

Meta pages the ad library by cursor, and the first page is a different size from the rest: 30 ads on page 1, then 10 per cursor page.

from scavio import ScavioClient

client = ScavioClient()

ads, cursor = [], None
while True:
    # page_id is the advertiser's numeric Facebook Page id, as a string.
    page = client.meta_ads.advertiser("20531316728", cursor=cursor)["data"]
    ads.extend(page["ads"])
    if not page["has_next_page"]:
        break
    cursor = page["next_cursor"]

print(len(ads), "ads")

# Every page is 1 credit, and past the first 30 ads a page is 10 ads, so a deep
# walk costs roughly one credit per ten ads. Budget for depth.

# Keyword search pages the same way. total_results caps at 50000 with
# total_is_capped true, because Meta itself only reports ">50,000" -- never
# present it as an exact count.
first = client.meta_ads.search(
    "black friday", country="US", active_status="active"
)["data"]
if first["total_is_capped"]:
    print("more than", first["total_results"], "ads match")

# Spend, reach, impressions and the paid-for-by disclosure are political-only.
# On commercial ads they are null by design, not a bug.
political = client.meta_ads.search("vote", ad_type="political_and_issue_ads")["data"]

# The cursor is a self-contained blob: every other filter is ignored while it
# is set, so do not "change the filter and keep paging" -- start a new walk.

20. Costco -- Warehouse Prices, Stock, Gas, Coupons and Clearance

costco covers costco.com and costco.ca (country="ca"); search, product and warehouses also take the international sites (uk, au, mx, jp, kr, tw). Warehouse numbers come from warehouses(); pass one as warehouse_id to add that warehouse's in-store price, stock status and price code to every search, category or deals result.

from scavio import ScavioClient

client = ScavioClient()

# Nearest warehouses to a US zip code, with hours, services and gas prices.
near = client.costco.warehouses(zip="10025", limit=3)["data"]

# Search with one warehouse's in-store price alongside the online price.
oil = client.costco.search("olive oil", warehouse_id="1062", page_size=5)["data"]
for item in oil["results"]:
    print(item["title"], item["price"], item["warehouse"]["price"])

# Online vs in-warehouse price at up to 10 warehouses, with discount, final
# price and promotion dates. 1 credit for up to 9 warehouses, 2 for 10.
prices = client.costco.prices(item_id="1492456", warehouse_ids=["1062", "1107"])["data"]
for row in prices["items"][0]["warehouses"]:
    print(row["warehouse_id"], row["price"], row["final_price"])

# In-warehouse stock status (a status, never a quantity). 1 credit per 5 warehouses.
stock = client.costco.availability("1492456", ["1062"])["data"]

# Regular, premium and diesel prices near a location, or at given warehouses.
gas = client.costco.gas(zip="90001")["data"]
for station in gas["stations"]:
    print(station["name"], station["prices"]["regular"], station["prices"]["premium"])

# The current member coupon book (US), and the deal feeds.
book = client.costco.coupons()["data"]
hunt = client.costco.deals("treasure_hunt")["data"]

# Markdown finder at one warehouse: .97 clearance and .00/.88 manager markdowns
# by default; add special_buy for .49/.79/.89 endings. The codes are the
# member-community decode -- Costco does not publish them.
marked_down = client.costco.clearance("1062", query="snacks")["data"]
print(marked_down["count"], "of", marked_down["products_scanned"], "items marked down")

21. Trustpilot -- Business Search, Profiles, Reviews and Categories

trustpilot finds businesses by keyword or category, reads a full profile (TrustScore, rating distribution, reply rate and days to reply on negative reviews, AI summary and topics when Trustpilot has them), and pages filtered reviews. Every call costs 2 credits. business() and reviews() take exactly one of domain or url.

from scavio import ScavioClient

client = ScavioClient()

# Businesses matching a keyword, up to 100 per page. Email and phone appear
# only when the business published them.
found = client.trustpilot.search("vpn", country="US", page_size=10)["data"]
for biz in found["businesses"]:
    print(biz["name"], biz["domain"], biz["trust_score"], biz["review_count"], biz["email"])

# One profile with the rating distribution, reply behavior and 20 newest reviews.
profile = client.trustpilot.business(domain="nordvpn.com")["data"]
print(profile["trust_score"], profile["rating_distribution"])
print(profile["reply_behavior"]["reply_rate"], profile["reply_behavior"]["avg_days_to_reply"])
topic_ids = [t["id"] for t in profile["topics"]]  # feed these to reviews(topics=...)

# Filtered reviews, 20 per page. Trustpilot serves up to 10 pages (200 reviews)
# per filter combination; max_page says how far this one goes. Slice by stars,
# language, date_range, topics or search to reach more. total_filtered counts
# every match, so a date range doubles as review velocity.
bad = client.trustpilot.reviews(
    domain="nordvpn.com", stars=[1, 2], language="en", date_range="last30days"
)["data"]
print(bad["total_filtered"], "1-2 star reviews in 30 days; pages:", bad["max_page"])
for review in bad["reviews"]:
    print(review["rating"], review["title"], (review["reply"] or {}).get("text"))

# The category tree, then ranked businesses in one category. min_trust_score is
# a star-rating floor: Trustpilot rounds TrustScore to stars, so 4 includes 3.8+.
tree = client.trustpilot.categories()["data"]
print(tree["count"], "top-level,", tree["subcategory_count"], "subcategories")
ranked = client.trustpilot.category(
    "vpn_service", sort="reviews_count", min_trust_score=4, claimed_only=True
)["data"]
for biz in ranked["businesses"]:
    print(biz["name"], biz["trust_score"], biz["website"])

# One review by id, with the business it is about.
one = client.trustpilot.review("6aca2504b52cc4b518bf4b75")["data"]
print(one["review"]["text"], one["business"]["name"])

An unknown business, review or category returns a 404 (NotFoundError) and is billed 2 credits. Trustpilot keeps no TrustScore history, so none is returned.

22. Realtor.com and Rightmove -- Listings, Sold Comps and Agents

realtor covers US listings (for sale, pending, for rent and sold) with up to 200 rows per call, full property detail, market stats and agents. rightmove covers UK listings, England and Wales sold house prices since 1995 and agent branches. Every call on both costs 1 credit.

Docs: Realtor.com search, property, locations, market, agents, agent; overview at the Realtor.com API page. Rightmove search, property, sold, sold property, locations, agents, agent; overview at the Rightmove API page.

from scavio import ScavioClient

client = ScavioClient()

# Sold land comps: 200 rows per call, filtered by lot size, price and recency.
# Realtor.com serves the first 10,000 results of a search; slice by price band,
# property type or area to go further. Sold prices are null in US non-disclosure
# states (Texas, Utah, Wyoming and others); sold dates still come back.
comps = client.realtor.search(
    location="Maricopa County, AZ",
    listing_status="sold",
    property_types=["land"],
    sold_within_days=365,
    lot_sqft_min=43560,
    page_size=200,
)["data"]
print(comps["total"], "sold; pages:", comps["max_page"])
for row in comps["listings"]:
    print(row["sold_price"], row["sold_date"], row["lot_sqft"], row["address"]["line"])

# Listing-text keyword watch, then one property in full. A listing must contain
# ALL the keywords you pass, so each extra keyword narrows the results.
shops = client.realtor.search(location="Austin, TX", keywords=["workshop"])["data"]
detail = client.realtor.property(property_id=shops["listings"][0]["property_id"])["data"]
print(detail["list_price"], detail["risk"]["flood_factor"], len(detail["property_history"]))

# Autocomplete -> market stats, and agents by ZIP code.
slug = client.realtor.locations("Austin")["data"]["locations"][0]["slug_id"]
market = client.realtor.market(slug)["data"]
print(market["median_listing_price"], market["median_days_on_market"])
agents = client.realtor.agents(postal_code="78744")["data"]
profile = client.realtor.agent(agents["agents"][0]["agent_id"])["data"]
print(profile["name"], profile["rating"], len(profile["reviews"]))

# Rightmove: resolve a place, search for sale (24-25 per page, up to 42 pages).
# There is no keyword filter; match on summary and key_features yourself.
loc = client.rightmove.locations("Bristol")["data"]["locations"][0]["location_id"]
homes = client.rightmove.search(location_id=loc, property_types=["terraced"], max_price=350000)
for home in homes["data"]["listings"]:
    print(home["price_display"], home["address"], home["tenure"])

# Sold prices since 1995 for an outcode (England and Wales only), then one
# address's full history.
sold = client.rightmove.sold("BS3", sold_within_years=5)["data"]
print(sold["summary"]["average_price_last_year"])
history = client.rightmove.sold_property(sold_id=sold["properties"][0]["sold_id"])["data"]
for sale in history["transactions"]:
    print(sale["date"], sale["price_display"], sale["tenure"], sale["percent_change"])

# Agent branches in an area, then one branch.
branches = client.rightmove.agents(location="Bristol")["data"]
branch = client.rightmove.agent(branch_id=branches["agents"][0]["branch_id"])["data"]
print(branch["name"], branch["listings"]["for_sale_count"], branch["listings"]["to_rent_count"])

An unknown property, listing, sold id, agent or branch returns a 404 (NotFoundError) and is billed 1 credit; a removed Rightmove listing is also a 404. Rightmove serves 42 search pages, 40 sold pages and 50 agent pages per query.

23. Mercado Libre -- Listings, Products, Reviews, Questions and Best Sellers

mercadolibre covers Mercado Libre in Mexico, Argentina and Colombia and Mercado Livre in Brazil (country="mx", "br", "ar" or "co"; prices come back in MXN, BRL, ARS or COP). reviews costs 1 credit; search, product, questions and best_sellers cost 2. Chile is not supported.

Docs: Mercado Libre search, product, reviews, questions, best sellers; overview at the Mercado Libre API page.

from scavio import ScavioClient

client = ScavioClient()

# Search by keyword (about 48 rows per page, up to 42 pages, so about 2,000
# results per query; slice by price range, condition or seller to reach more).
found = client.mercadolibre.search(
    "iphone 15",
    country="mx",
    free_shipping=True,
    sort="price_asc",
)["data"]
print(found["total"], found["auto_applied_filters"])
for row in found["results"]:
    # sold_quantity_min is a lower bound ("+1000 vendidos"), never an exact count.
    print(row["price"], row["original_price"], row["sold_quantity_min"], row["title"])

# One catalog product: price, buy-box seller with reputation, star distribution.
product = client.mercadolibre.product(product_id="MLM1027172667")["data"]
print(product["price"], product["seller"]["nickname"], product["rating"]["distribution"])

# A seller's catalogue: the buy-box seller_id feeds a filtered search.
catalogue = client.mercadolibre.search(
    country="mx", seller_id=str(product["seller"]["seller_id"]), condition="new"
)["data"]

# Reviews (15 per page, 1 credit), e.g. only 1-star; original_text keeps the
# untranslated review and country says where it was written.
reviews = client.mercadolibre.reviews(product_id="MLM1027172667", rating=1)["data"]
for review in reviews["reviews"]:
    print(review["rating"], review["country"], review["text"])

# Buyer questions with the seller's answers; page with next_offset.
qa = client.mercadolibre.questions(product["item_id"])["data"]
for q in qa["questions"]:
    print(q["text"], "->", q["answer"]["text"] if q["answer"] else None)

# Top 20 best sellers in a category (mx, ar or co; not Brazil). The category
# prefix must match the country.
top = client.mercadolibre.best_sellers(product["category_id"], country="mx")["data"]
for row in top["results"]:
    print(row["rank"], row["price"], row["sold_label"], row["title"])

A zero-result search returns an empty page and is billed. An unknown product, item or seller nickname returns a 404 (NotFoundError) and is billed, as does a listing with no public questions. Only the buy-box seller is returned per product, and Brazil and Argentina may auto-apply a "new" condition (listed in auto_applied_filters).

Error Handling

from scavio import (
    ScavioClient,
    InvalidAPIKeyError,
    RateLimitError,
    InsufficientCreditsError,
    NotFoundError,
    BadRequestError,
    ScavioConnectionError,
    ScavioTimeoutError,
    ScavioAPIError,
    ScavioError,
)

client = ScavioClient(api_key="sk_...")

try:
    results = client.search("query")
except InvalidAPIKeyError:
    print("Check your API key")
except RateLimitError:
    print("Too many requests - upgrade your plan")
except InsufficientCreditsError:
    print("Out of credits - purchase more at dashboard.scavio.dev")
except ScavioAPIError as e:
    # Any other non-2xx response; inspect the details:
    print(e.status_code, e.response_body)

All exceptions inherit from ScavioError. HTTP errors (BadRequestError 400, InvalidAPIKeyError 401, InsufficientCreditsError 402, NotFoundError 404, RateLimitError 429, ScavioAPIError for anything else) carry .status_code and .response_body. Network failures raise ScavioConnectionError / ScavioTimeoutError after retries are exhausted.

Configuration

client = ScavioClient(
    api_key="sk_...",
    base_url="https://api.scavio.dev",  # custom base URL
    timeout=30.0,                        # request timeout in seconds
    max_requests_per_second=1,           # client-side rate limit (1-50)
    max_retries=2,                       # retries on 429/5xx/network (0 disables)
)

Async client

The async client mirrors the sync one method-for-method. It keeps a single pooled httpx.AsyncClient alive for its lifetime; close it with await client.aclose() or use the async context manager.

import asyncio
from scavio import AsyncScavioClient

async def main():
    async with AsyncScavioClient(api_key="sk_...") as client:
        return await client.google.search("openai", gl="us")

asyncio.run(main())

Integrations

Scavio works with popular AI/LLM frameworks:

• LangChain -- pip install langchain-scavio
• MCP Server -- for Claude, Cursor, and other MCP clients
• n8n -- no-code workflow automation

API Reference

Every method's full parameter list is available inline in your editor (typed keyword arguments with docstrings). See the API docs for field-level details.

Changed in 0.22.0

• 19 new methods across TikTok, YouTube, Reddit and X, 1 credit each.
  • tiktok: trending_searches(), trending_videos() (about 6 videos per call whatever count says), search_suggestions(keyword), music(music_id), music_videos(music_id) (count is ignored upstream, about 14 per page), search_music(keyword), music_chart() (Top 50 or Viral 50, one global chart), user_similar(sec_user_id), user_playlists(sec_user_id), playlist_videos(playlist_id), user_live(username) (an unknown username is a billed 404) and user_region(username) (an unknown username is an unbilled 404; also returns sec_user_id). These return TikTok's raw payload: large numeric ids can lose precision as JSON numbers, so use id_str / sec_uid.
  • youtube: community_post(post_id), community_post_comments(post_id).
  • reddit: search_comments(query) (sort RELEVANCE, HOT, TOP or NEW), search_subreddits(query).
  • x: search_communities(keyword), community(community_id), community_tweets(community_id) (ranking Relevance or Recency).
• youtube.transcript now costs 5 credits (was 8); the docstring says so.

chart = client.tiktok.music_chart(chart="viral", count=10)
region = client.tiktok.user_region("mrbeast")
similar = client.tiktok.user_similar(region["data"]["sec_user_id"])
threads = client.reddit.search_comments("web scraping api", sort="NEW")
posts = client.x.community_tweets("1508567430188126209", ranking="Recency")

Changed in 0.21.1

• README: Mercado Libre docs and landing page links; wording cleanup. No behavior changes.

Changed in 0.21.0

• New mercadolibre namespace, 5 methods. search (keyword or seller in Mexico, Brazil, Argentina or Colombia; condition, price, discount, free shipping, Full fulfillment, official store and interest-free installment filters; relevance or price sort; 2 credits), product (buy-box seller, variations, star distribution, specs; 2), reviews (15 per page with a star filter; 1), questions (buyer questions with seller answers; 2) and best_sellers (top 20 in a category, not Brazil; 2). country is required on search and best_sellers and is keyword-only. See example 23 above.
• realtor.search(keywords=...) docstring: a listing must contain ALL the keywords, so each extra keyword narrows the results.

Changed in 0.20.1

• README: Realtor.com and Rightmove docs links; Rightmove sold prices cover England and Wales only, and search pages return 24-25 rows (the rightmove.search(page=...) docstring now says so). No behavior changes.

Changed in 0.20.0

• New realtor namespace, 6 methods, 1 credit each. search (for sale, pending, for rent or sold by location, radius or agent; price, beds, baths, sqft, lot, garage, HOA, year built, property type, recency, listing-text keywords and more; up to 200 rows per call), property (price and tax history, valuations, schools, risk, listing agent), locations, market, agents and agent.
• New rightmove namespace, 7 methods, 1 credit each. locations, search (buy or rent by location, place name or branch), property, sold (sale history since 1995 per address plus an area summary), sold_property, agents and agent. See example 22 above.

Changed in 0.19.1

• trustpilot.category(min_trust_score=...) docstring corrected: the filter is a minimum star rating (3, 4 or 4.5). Trustpilot rounds TrustScore to stars, so 4 includes TrustScore 3.8 and up. No code changes.

Changed in 0.19.0

• New trustpilot namespace, 6 methods, 2 credits each. search (businesses by keyword, up to 100 per page), business (full profile by domain or Trustpilot URL), reviews (stars, language, date range, sort, verified-only, with-replies, topics and text filters; up to 200 per filter combination), categories (the tree or a name lookup), category (ranked businesses with sort, minimum star rating and claimed-only) and review (one review by id). See example 21 above.

Changed in 0.18.1

• Documentation links only; no code changes. The client already identifies itself on every call with User-Agent: scavio-python/<version> and X-Client-Source: scavio-python.

Changed in 0.18.0

• New costco namespace, 13 methods. search, category, categories, product (up to 20 ids per call), prices (online vs in-warehouse price at up to 10 warehouses), availability (in-warehouse stock status), reviews, warehouses, gas, coupons (the member coupon book), deals (new, while_supplies_last, treasure_hunt, member_favorites, online_only, on_sale), clearance (price-code markdown finder) and autocomplete. See example 20 above.

Changed in 0.17.0

• Walmart store targeting. walmart.stores(zipcode, domain=...) is new: it returns the Walmart stores near a US ZIP (walmart.com) or a Canadian postal code (domain="ca", walmart.ca), nearest first, each with its store_id. Pass that store_id together with delivery_zip to walmart.search() or walmart.product() on the same domain to get that store's assortment and availability; data.location echoes the store used. Store-targeted calls cost 2 credits and take 10-60 seconds, so give them a generous timeout. walmart.product() also accepts domain="ca", together with a store pair.

  stores = client.walmart.stores("50036")
  store_id = stores["data"]["stores"][0]["store_id"]  # "1389", Boone Supercenter
  milk = client.walmart.search("whole milk", delivery_zip="50036", store_id=store_id)
  print(milk["data"]["location"])  # {"city": "Boone", "state": "IA", "store_id": "1389", ...}
  

• delivery_zip and store_id are no longer retired; device is now the only retired Walmart parameter.

Changed in 0.16.0

• max_requests_per_second now accepts up to 50 (it was capped at 10), matching the raised per-plan request limits -- the top plan allows 50, so the old ceiling rejected a value the plan entitled you to. The default is still 1, and the bound now lives in one place (scavio._ratelimit.MAX_REQUESTS_PER_SECOND) shared by the sync and async clients.

Changed in 0.15.0

• 21 new namespaces with 85 endpoints: Threads, Kuaishou, eBay, Target, Home Depot, Zillow, Redfin, Booking.com, Tripadvisor, Airbnb, Yelp, Indeed, Glassdoor, Apple App Store, Google Play, SEC EDGAR, Companies House, G2, Capterra, Google Ads Transparency and the Meta Ad Library. With the Walmart rebuild and extract below, that is 93 endpoints in this batch and 195 in the SDK. Every method is typed and documented inline, on both the sync and async clients.
• client.extract(url, format=..., mode=...) is new, and is a top-level method, not a namespace. Reading an arbitrary URL is a core capability rather than a platform, so it is client.extract(...), never client.extract.extract(...).
• Walmart grew from 2 endpoints to 7 (reviews, category, offers, seller, seller_products are new) and search/product changed shape. device, delivery_zip and store_id are retired: the API answers 200 with a top-level warnings array if you send them through **extra. domain (com | ca | com.mx) is new on search and category, and it is the price-bearing parameter. page supersedes start_page, which stays as a deprecated alias.
• Walmart min_price / max_price widened from int to float. The backend was always z.number(), so 19.99 was always accepted; the old annotation was simply wrong. No runtime behaviour changed.
• Four surfaces are body-priced and their docstrings say so instead of quoting a flat price: Walmart (com.mx costs 2), Threads (4 credits when addressed by username instead of user_id), Kuaishou (1, 2, 10 or 40 per endpoint) and extract (mode="ultra" costs 2).
• Some caveats worth knowing before you write a loop: eBay sold=True returns total_results: null; Meta Ad Library pages are 30 ads then 10, and the cursor ignores every other filter; Capterra search does not paginate at all; Apple App Store search has no pagination either (raise limit, up to 200); Airbnb prices exist only on search, never on listing.

Changed in 0.14.0

• reddit.post() now takes post_id as well as url -- pass either one. url stays the first positional argument, so existing calls are unaffected. The response is a flat post object and carries no comments; use reddit.post_comments() for those.
• youtube.shorts(sort_by=...) is now typed as relevance | date | view_count | rating instead of a free-form string.
• youtube.search() lost its location flag. It was never part of the backend schema, so it was silently dropped rather than filtering anything.

Amazon changed in 0.12.0 (breaking)

The Amazon API now returns a normalized shape instead of the previous raw payload.

• search returns {query, page, total_results, total_results_text, count, products[], filters[], related_searches[]}. Each product is {asin, title, url, image, price, currency, rating, reviews_count, is_sponsored, position, badge, sales_volume, delivery{is_free, date, fastest_date}}.
• product returns flat fields: price, list_price, currency, rating, reviews_count, features, images, videos, variants, specifications, best_sellers_rank, shipping, and more. The old buybox[] array no longer exists -- use offers for per-seller pricing.
• offers is new: every seller for one ASIN, with price, condition, seller_name, is_buy_box_winner, is_fulfilled_by_amazon, and delivery windows.
• country (ISO 3166-1 alpha-2: us, gb, de) is the marketplace selector and replaces domain. page replaces start_page. The old names still work as deprecated aliases.
• Nine parameters were removed: language, currency, device, sort_by, pages, category_id, merchant_id, zip_code, autoselect_variant. sort_by in particular was verified to be ignored by the marketplace, so result sorting is not available at any layer. Sending one of them anyway (via **extra) still returns 200, with a top-level warnings array explaining what was ignored.
• options still returns domains and countries; languages and currencies are now always empty, because neither is a request parameter any more.

Links

• Website
• Documentation
• Dashboard & API Keys
• API Reference
• Compare Scavio vs alternatives

License

MIT

About Scavio

Scavio is a unified search API built for AI agents — one API key, structured JSON. A real-time Tavily alternative and SerpAPI alternative with data from:

• Google Search API — SERP results, news, images, maps, and knowledge graph
• Amazon Product API and Walmart Product API — product search and details
• YouTube API, TikTok API, and Instagram API — video and social media data
• TikTok Shop API — product search, detail, reviews, categories, and shop catalogs
• Reddit API — posts, comments, subreddits, and trending
• X API and LinkedIn API — tweets, profiles, companies, and jobs

For a detailed head-to-head breakdown, see Tavily vs Scavio.

Get a free API key and explore the documentation.