google.search, belong to the per-site API that API v1 replaced.
Lower prices, costUsd on webhooks, a tool for each product on MCP, and GitHub search and Google Scholar removed
New
- Job webhooks (crawl, batch, agent and extract) and
monitor.check.completednow carrycostUsd, the exact dollars charged.creditsUsedandactualCreditsstay as older fields, rounded up to whole $0.002 steps. See Webhooks and Monitor webhooks.
-
Lower prices, from November 18. The list prices dropped. Per 1,000 unless noted, before and after:
- A page read in scrape, crawl, batch scrape, search page text and monitor checks, and most parser calls: $2 to $1.
- Map: $2 to $0.50.
- Search, for each 10 results from each source: $0.004 to $0.001.
jsonwith a schema or a prompt,summary,highlights,question,product,menu,changeTrackinginjsonmode and each agent step: $8 to $3 more.- Crawl preview: $8 to $3.
checkPromptInjection: $8 to $2.50 more.- Extract: $10 to $5 for most pages, so one page costs $0.005 instead of $0.01.
- Parse and PDF text from a text layer: $2 to $0.50. Text read from a picture: $2 to $2.50.
- Parsers that charge for each 10 results (reviews, apps, listings, videos): $0.002 to $0.001. Reddit comments, for each 200: $0.002 to $0.001.
- Monitors:
jsoncomparison $8 to $3 more, a judged change $2 to $1 more. - These did not change: browser sessions and tasks, session traffic at $5 per GB, Instagram and TikTok audio transcripts,
zeroDataRetention, andzillow-search.
-
The MCP server lists a tool for each product. Without an API key it lists
stophy_search,stophy_scrape,stophy_map,stophy_parseandstophy_signup. With one it adds tools for crawl, extract, agent, browsers, developer and paper search, and monitors, such asstophy_crawl_statusandstophy_monitor_create.?toolsets=adds tools for more endpoints of a source. See Install MCP.
- The MCP tools
stophy_search_endpoints,stophy_describe_endpointandstophy_callare gone. Use the tool for the product, orstophy_scrapewith a parser informatsfor a site’s data. - GitHub search is gone.
GETandPOST /v1/search/research/githubanswer410. Search READMEs, issues, pull requests and releases with developer search. - The
google-scholarparser is gone. Read papers withcategories: ["research"]on search, orPOST /v1/search/research/papers.
- A scrape with
maxAgewhose page has no stored copy reads the page at once. Before, it waited several seconds first. AminAgecall with no stored copy answersnotFoundat once. See Freshness and caching.
- With an API key, pages on some government, bank and payment sites that failed with
sourceUnavailablenow read.
New
- Billing is in dollars. On November 15, balances, usage, history and logs switched to dollars. Every response carries
costUsdand anx-cost-usdheader,GET /v1/usagereturnsbalanceUsd, and a top-up adds exactly the dollars paid. Prices did not change.creditsUsedstays and counts the same cost in steps of $0.002. See Pricing. - Save allow and block rules by domain with
POST /v1/browser/network-policies, then open a session withnetworkPolicyId. A change to the policy reaches running sessions within seconds. See Network policies. - Upload a Chrome extension with
POST /v1/browser/extensionsand load up to 3 in a session withextensionIds. See Extensions. - Send
recording: trueto record a session as an MP4, then read it withGET /v1/browser/{session_id}/recording. Videos are kept 7 days. See Recordings and replay. - Change how an open session reaches the web, with its tabs and logins kept, using
PATCH /v1/browser/{session_id}/proxy. Traffic costs $5 per GB by default and $0.20 per GB withproxy: "direct". See Network location. - Put a file in a session with
POST /v1/browser/{session_id}/uploads, and list what a session downloaded withGET /v1/browser/{session_id}/downloads. See Files. - Sessions take
screen,metadata,location,captureConsoleandsolveCaptchas, andttlgoes up to 6 hours. Free accounts keep 3 sessions open, paid accounts 10. See Browser sessions. - The MCP server opens, drives and closes a browser session with
browserOpen,browserRunandbrowserClose. See Install MCP. - Scrape takes
queryto return the passages of a long page that match it, andoffsetwithnextOffsetto read on pastmaxChars. See Content filtering. - Search takes
sbd:1intbsto sort newest first, andimagesize:WIDTHxHEIGHTinqueryto keep images of one size. See Search filters. - Free calls list your crawls and batches, page their results, delete a finished batch’s results, and read webhook deliveries with their attempts and a retry. See Webhooks.
- New accounts start with $2, down from $10. Sign-in is with Google or GitHub only. Existing balances stay.
- Each endpoint has its own limit: 100 calls a minute until your first top-up, then 600. Job starts, calls at once, waiting jobs and open sessions have limits by tier too, and
GET /v1/limitsshows yours. See Rate limits. - An answer from a stored copy costs the same as a live read, and a stored copy older than
maxAgeis never returned, even when the site fails. See Freshness and caching. - A scrape keeps trying for 60 seconds by default, and
timeouttakes 1 second to 5 minutes. See Timeouts. - A web result’s
descriptionis the result’s search snippet. Sendhighlights: trueto get the passages that best match your query instead. See Search page content.
- A call whose caller hangs up before the answer is refunded.
- A browser task runs only the whole minutes its
maxCostbuys. - A crawl keeps reading until it reaches its
limit. A large crawl could end early and reportcompleted. - A refused browser session sends
Retry-After. - A URL with a username or password is refused with
invalidRequest.
- The
qdr:hpast-hour value intbsis refused. The smallest window isqdr:d. - The
companiesandpeoplesearch categories,/v1/search/companies,/v1/search/peopleand thecompanyparser are gone. /v1/search/govis gone.
New
- API v1. On November 5, Stophy became
scrape,search,map,crawl,batch/scrape,extractandparse, withagent,monitorandbrowserbeside them. Every answer is{ success, data }with arequestId, and every error is flat. See the Introduction. - Parsers. Each site is a parser. Send its page address to scrape with
{ "type": "json", "parser": "reddit-search" }, andGET /v1/parserslists about 100 with their input and an example answer. See Pre-built parsers. - Scrape formats:
summary,question,product,menu,branding,rawBase64,audioandvideofor YouTube links,jsonwith aschema, apromptor CSSrules, andattributes. See Scrape. - Scrape options:
actionsup to 50 steps,profile,lockdown,checkPromptInjection,redactPII,onlyCleanContentandzeroDataRetention.GET /v1/scrape/{id}reads a scrape again for 24 hours, andPOST /v1/scrape/{id}/interactkeeps working on the page. See Browser actions and Interact. POST /v1/crawlandPOST /v1/batch/scrapereturn any scrape format, report failed pages at/errors, and take awebhook. A batch takesappendToIdto add pages while it runs, andGET /v1/crawl/activelists running crawls. A job over your running limit waits in line. See Crawl and Batches.POST /v1/extractfills the fields you name from up to 25 pages, each checked against the page, for $10 per 1,000 pages. A call that finds none of them is free.POST /v1/agentsearches, reads pages and fills your fields with a source for each. See Extract and Agent.POST /v1/monitorwatches pages on a schedule and reports what changed. Webhooks are signed withStophy-Signature, andPOST /v1/webhook/secretreads or replaces the secret. See Monitors.POST /v1/parseturns a PDF, Word, Excel, PowerPoint or HTML file, or a picture of text, into Markdown. Scanned PDF pages are read,POST /v1/parse/upload-urltakes large files, andGET /v1/parse/formatslists the types. See Parse documents.- Search takes
categoriesofresearch,pdfandvideos, and hasPOST /v1/search/research/papersandPOST /v1/search/developer. It also had company, people and government search, which were removed on November 14. See Search. - Browser sessions:
POST /v1/browseropens one for $0.06 per hour,POST /v1/browser/{session_id}/executeruns code or a prompt in it, and profiles keep logins.POST /v1/browser/taskclicks through a public site toward a goal. See Browser sessions and Browser tasks. - An agent signs up with
POST /v1/signup, a person approves, andPOST /v1/signup/pollreturns the key once. See API keys. GET /v1/limitsshows your tier, balance, rate limit and key expiry.GET /v1/statusshows which sources passed their last checks.POST /v1/feedbackrates a call, and a bad rating refunds what it cost, up to a daily cap.- A repeated
POSTwith the sameIdempotency-Keyreturns the first answer for 24 hours. - Search, scrape, map and parse work without an API key, up to 10 calls a minute and 50 calls a day per IP address. See Rate limits.
- On November 5, new accounts start with $10 of free credit, up from $5.
- Every error says what to do next. Errors carry
error,code,retryableandrequestId. A wrong method is405 methodNotAllowed, and a key that does not work is401 unauthorized. See Errors.
- The per-site endpoints, such as
/v1/google/searchand/v1/youtube/transcript, are replaced by parsers, and the old paths answer404. The AI search, keyword suggestion and AI answer endpoints are removed. - Scrape takes one
url. A call withurls, added on October 27, is refused and namesPOST /v1/batch/scrape.
- The TypeScript and Python SDKs, both
stophy1.3.0, call API v1. The site methods are gone. Read a site withparsed(parser, url, options). See SDKs. - The CLI,
@stophy/cli1.3.0 and 1.4.0, works with API v1. In an agent or a pipe it prints one line of JSON, and without an API key it starts a signup for a person to approve. See Install the CLI. - The hosted MCP server,
@stophy/mcp1.3.1, hasstophy_search,stophy_scrape,stophy_map,stophy_crawl,stophy_extractandstophy_signup. A site’s data comes fromstophy_scrapewith a parser informats. - The agent skills call API v1, one skill per task. See Install skills.
New
- Craigslist search and listings, Luma events, Realtor.com search and properties, and Expedia hotels. Today they are the
craigslist-search,craigslist-listing,luma-events,luma-event,realtor-search,realtor-propertyandexpedia-hotelsparsers. web.scrapetakesurls, 2 to 10 pages in one call, and returnsresultsin the order you sent. API v1 replaced this withPOST /v1/batch/scrape.web.searchranks fresher, cleaner results first and filters by language and date. Today those arelangandtbson search.- Google Flights takes a currency and returns the link to its search page. See the
google-flightsparser. - A scrape says when it left out network calls or cropped a screenshot, in
networkDroppedandscreenshotCropped. See Network.
- Search with
includeDomainsreturns the on-site results, also withsafe.
- Reddit search, posts and user pages read again after a change on the site.
- A page that never finishes loading, or that is only a loading notice, costs nothing.
New
- The
highlightsformat returns the passages of a page that best match aquery, each with its section heading. See Highlights. - The
changeTrackingformat says what changed since your key last read a page. See Change tracking. - The
networkformat returns the JSON a page loaded while it rendered. See Network. linkStylewrites linksinline,numberedornone. See Content filtering.- A
screenshotcomes back as a link instead of image data. See Screenshot. - A crawl with a
promptfollows link text, not only addresses. See Crawl scope.
- Main content leaves out blocks that are mostly links and blocks a site repeats on every page.
- A page that only says it needs JavaScript now comes back with its content.
web.mapreturns its pages in the same order on every call.
- A call that fails halfway gives its charge back, and a refunded or disputed top-up takes its balance back.
- A field of the wrong type fails the call with
invalidRequestinstead of goingnull.
New
- Scrape, map, parse, crawl and batch are live. Today they are
POST /v1/scrape,/v1/map,/v1/parse,/v1/crawland/v1/batch/scrape. - Web, news and image search work without an API key, within the free limit. See Rate limits.
- X profile posts, newest first or X’s popular set. Today that is the
x-profileparser. - Amazon reviews, newest first, by star rating and verified purchases only. Today that is the
amazon-reviewsparser. web.contactsreads a site’s emails, phone numbers and social links. Today theemailsandphone-numbersparsers read emails and phone numbers.
- Pinterest pins no longer carry counts that Pinterest stopped sending.
google.jobsandgoogle.aiModeon October 13, andgoogle.shoppingon October 16.
New
- Airbnb, Booking.com, company career pages, Threads, Trustpilot and eBay. Today they are parsers, such as
airbnb-search,booking-search,careers-jobs,threads-post,trustpilot-companyandebay-search. - Facebook pages, posts, comments and Marketplace, Yelp search, businesses and reviews, and Glassdoor company reviews and salaries. Today they are parsers such as
facebook-page,yelp-searchandglassdoor-reviews. - Nine more endpoints work without an API key, within the free limit.
- Request fields use the same names as the fields in the response, each endpoint takes the filters its site has, and one rule sets the price of every call.
- Long text comes back whole. Add
?maxTextLength=<characters>to the URL to cut it.
- Pages of results stop repeating rows, and a cursor Stophy did not issue is refused before any call.
- A search with nothing relevant answers “no results” at once.
- Amazon best sellers fill all 50 ranks.
- 11 rarely used endpoints: YouTube charts, related videos, hashtags and channel search, Google videos, Reddit domain, Microsoft ads search, ad and advertisers, and Pinterest ads search and ad.