A striking number of the small APIs and free tools that developers quietly built on over the last decade are being shut down, paywalled, or restricted in 2025–2026. Google is closing its Custom Search JSON API to new customers, its cache: operator is gone, Dark Sky’s weather API has been retired, goo.gl links stopped resolving, Alexa’s rank data is dead, Heroku removed its free tier, and a long tail of “free” data endpoints have added keys, quotas, or paywalls. If you built a workflow on any of them, you’ve either already felt it or you’re about to.
This is a single, maintained reference for that whole pattern: what changed, what to use instead, and — where it makes sense — how to replace a dead endpoint with a pay-per-use feed you control instead of another free API that might disappear next year. It replaces a series of individual posts we published on each of these migrations; consolidating them into one page you can search is more useful than a dozen you have to find. Each section is self-contained, so jump to the one that broke your build.
Web search: Google Custom Search JSON API is closing to new customers
Google’s Programmable Search Engine / Custom Search JSON API — the endpoint a lot of “search this site/the web” features were built on — is closing to new customers on January 1, 2027, and the free 100 queries/day tier has always been narrow. Existing projects keep working for now, but new sign-ups are being cut off, which makes it a dead end to build on.
What to use instead. For genuine web-search results, a maintained SERP API (Bing’s, or a scraping-based SERP feed) is the durable replacement; for site search, a dedicated index (Algolia, Typesense, Meilisearch) beats bolting web search onto your own pages. If what you actually need is structured results you can store and re-query rather than live search-as-a-service, a pay-per-result SERP feed is more predictable than a quota you can hit at the worst moment. The migration itself is mechanical: swap the endpoint, map the response fields (title, link, snippet), and drop the per-day quota logic you no longer need.
Cached pages: the cache: operator and Google Cache are gone
Google removed cached-page links from search results and retired the cache: operator. Anything that relied on pulling Google’s cached copy of a page — for change detection, for reading a page that’s temporarily down, for archival — is broken.
What to use instead. The Internet Archive’s Wayback Machine and its availability API are the closest drop-in for “give me a recent copy of this URL,” and they’re free and stable. For your own change-detection needs, the honest fix is to stop relying on someone else’s cache and store your own snapshots on a schedule — fetch the page yourself, keep the HTML, diff it. That’s a few lines of code plus storage, and it never gets deprecated out from under you.
Site popularity: Alexa rank is dead — use Tranco
Amazon shut down Alexa.com in 2022, and the “Alexa rank” that a generation of SEO tools quoted no longer exists. The academic-grade replacement is Tranco, a research-maintained ranking of the top sites that’s specifically designed to be reproducible and hard to game — which is exactly why it’s a better input for any pipeline than Alexa ever was. It’s free to download as a list; there’s no per-call API to rate-limit, so you ingest the daily list and join against it locally.
Company & funding data: the Crunchbase free API is effectively gone
Crunchbase’s usable free/low-cost API tiers have narrowed to the point where most builders can’t rely on them. The good news is that the underlying data — who raised, from whom, when — is largely public at the source. Funding disclosures, company registrations, and news announcements are all available without paying an aggregator’s markup.
What to use instead. Pull from the primary sources directly: SEC Form D filings for US private raises, Companies House for UK raises, and funding-news feeds for the announcements. NexGenData’s Crunchbase News Scraper pulls the daily funding, M&A and IPO headlines as structured JSON, and UK Startup Funding Tracker detects UK rounds straight from Companies House SH01 filings — often on the public record before the press covers them. You pay per record instead of per seat, and the source is one you can audit.
Weather: Dark Sky is retired
Apple retired the Dark Sky API at the end of 2022 and folded it into WeatherKit. If you’re not in the Apple ecosystem, the durable free/low-cost replacements are Open-Meteo (free, no key, generous limits) and the national weather services’ own APIs (NOAA/NWS in the US). Migration is a field-mapping exercise — hourly/daily blocks map cleanly — with the one gotcha that minute-by-minute “hyperlocal” precipitation, Dark Sky’s signature feature, isn’t universally available; Open-Meteo covers most of the rest.
Technology fingerprinting: Wappalyzer went commercial
Wappalyzer’s move to paid API access left a gap for “what’s this site built with?” at scale. The fingerprint definitions themselves are open, and there are maintained open-source alternatives — webappanalyzer (the community fork of the rules) plus scanners like whatweb — that you can run yourself against a list of domains. If you’re doing this for lead-gen or competitive research, running your own fingerprinting over a domain list is cheaper and more flexible than a per-lookup API.
Domain intelligence: WHOIS is being replaced by RDAP
The old free-text WHOIS protocol is being phased out in favor of RDAP (Registration Data Access Protocol), which returns structured JSON instead of the inconsistent text blobs every WHOIS parser had to fight. If you’re still scraping WHOIS text, switching to RDAP removes an entire category of parsing bugs — registrar, dates, and status come back as clean fields. Registries publish RDAP endpoints directly, so bulk domain lookups don’t need a paid WHOIS reseller. For the adjacent “is this domain’s email/DNS set up correctly” question, NexGenData’s Domain Security Posture Checker returns DNS, SPF, DMARC and TLS status for a domain in one call.
IP geolocation: ipinfo and MaxMind tightened their free tiers
The free IP-geolocation tiers from ipinfo and MaxMind have gotten smaller and more restricted. For most use cases — coarse country/region-level geolocation for analytics or routing — you don’t need a paid tier: MaxMind’s GeoLite2 database is still downloadable and you can look up IPs locally at zero marginal cost, and open datasets cover the rest. Reserve the paid APIs for cases that genuinely need street-level accuracy or carrier data; everything else runs offline against a local database.
Short links: goo.gl stopped resolving
Google fully shut down the goo.gl URL shortener — existing goo.gl links no longer redirect. If you have historical data full of goo.gl links, the only way to recover their destinations is the ArchiveTeam / Internet Archive datasets, which captured a large share of goo.gl mappings before shutdown. The lesson for anything you control: don’t build on a third-party shortener you can’t export — run your own, or store the full destination alongside every short link.
Hosting: Heroku removed its free tier
Heroku’s elimination of free dynos in late 2022 pushed a lot of small projects to migrate. The mechanical part is easy (containerize, deploy elsewhere); the part people underestimate is cost modeling — a “free” side project can become a surprising monthly bill on some platforms. The durable free/cheap tiers today are Render, Railway, Fly.io, and Cloudflare Workers for edge functions; the right choice depends on whether you need a persistent process, a database, or just request/response. Model the cost at your real traffic before committing, not at zero.
Page-speed & site health: Lighthouse alternatives at scale
Lighthouse and PageSpeed Insights are excellent for a one-off audit but painful to run across many URLs on a schedule. For monitoring — catching a Core Web Vitals regression before users do — you want something you can point at a list of URLs and run daily. Options range from self-hosting Lighthouse CI to using a scheduled scanner that returns the same metrics as JSON per URL. The principle is the same as everything else on this page: turn a manual, one-URL tool into a scheduled feed you can diff over time.
The pattern — and how to migrate deliberately
Every situation above is the same shape: a convenient third-party endpoint disappears or paywalls, and the people who built on it scramble. The durable response isn’t to find the next free API that will die in two years — it’s to move one level closer to the primary source (the registry, the official dataset, the archive) and, where you need it as a live feed, run a scheduled job you control that returns clean JSON. That’s more work up front and far less fragile after.
A practical migration checklist: (1) identify the exact fields your code consumes from the dying endpoint; (2) find the primary source or a maintained alternative that carries those fields; (3) map the response and delete any quota/rate-limit handling you no longer need; (4) if it needs to be a live feed, schedule your own pull and store the results so you’re never dependent on someone else’s uptime; (5) keep a dated snapshot so a future deprecation is a data-migration, not an outage. NexGenData builds several of these replacement feeds as pay-per-use APIs on Apify — you can browse the full catalog and run any of them once against your own input before committing. This page is updated as more of these endpoints change; if one you rely on isn’t covered yet, it will be.
We cover AI agents, automation, and the tools that make them work. Our mission is to make AI accessible to everyone.