What is Supergood?

Record once, get an endpoint

You drive the workflow once in a cloud-hosted Chrome session. Supergood's homepage, read 8 September 2026, says the recording captures "page loads and network requests". An AI coding agent turns it into an integration, an OpenAPI spec and a customer guide, and you review the result as a diff before it ships. At runtime there is no browser. In the homepage's words: "Real APIs, not browser automation."

The clever part is that the agent reads more than it watched. In our 8 to 10 September 2026 field test, two recorded real-estate searches on a Polish court-auction portal were enough for it to pull the portal's whole taxonomy out of its compiled JavaScript: 8 real-estate and 17 movable sub-categories, and 16 provinces. A nationwide electronics search that nobody had demonstrated then returned HTTP 200 on its first live call. That is the difference between a macro recorder and something that reads a site's own source of truth.

Sign-ins run through a service account which, per the homepage, generates "a real email, phone number, and password". Every integration answers on POST /integrations/{slug}/{mode}, sync or async. For agents, the logged-in dashboard documents an MCP server with a service-account mode (seven tools, "60 req/15min") and an organization mode (eight tools, "100 req/15min").

It began by watching other people's API calls

Read that timeline like someone who has maintained scrapers and it explains the product better than the tagline does. Screen scrapers and RPA bots rarely fail on the day they are built. They fail weeks later, quietly, when a vendor moves a button, and the first alarm is somebody asking why the report is empty. Generating an unofficial API was never the hard half of the job; noticing that it broke was. A company that spent its early years watching third-party API calls in production is, on paper, built for that half. Klarfeld's 15 May 2026 post draws the same line: the first product showed that "network calls are deterministically observable and maintainable."

The Wayback Machine holds no readable capture of supergood.ai between 11 August 2025 and 11 August 2026, so the current tagline went live somewhere in that year.

What happened when we called it

On 15 September 2026 we deployed and called two integrations built during the field test. A US county parcel-records lookup answered 10 of 10 calls with HTTP 200, returning the same record as on 9 September, at a dashboard-displayed median of 490 ms (slowest 516 ms).

The auction integration returned HTTP 400 "Service account not found" 10 times out of 10, because we sent the documented example body with its placeholder account, [email protected]. The aftermath was the interesting part. API Logs listed all 20 calls, Anomalies put the auction integration at 100% failure, the Triage Bot escalated all 10, and each error was logged as a user error with the note an on-call engineer actually wants: "Retrying with the same args will fail identically."

One loose thread remains. That same note says the field "isn't even in the integration's strict request schema", while the published schema marks it required. Generated docs disagreeing with generated code is a small thing here, and one of the oldest integration bugs there is.

There is no price to quote

supergood.ai/pricing returned HTTP 404 on 9, 14 and 15 September 2026. On the last date the homepage's only pricing language was a call to action, "Need an integration to win a deal? Pay when you close", with no figure. Our Billing page read $0.00 "Not configured" on a demo-environment key, which says nothing about what a paying customer pays.

That button is worth reading closely. It speaks to software sellers who need an integration to close a deal, a different buyer from an operations team stuck with a portal. Work out which one you are before the sales call.

Who should buy it, and what will bite

Supergood makes sense for a short list of stubborn systems, such as portals, registries and vendor tools that will never ship an API, when you would rather own an endpoint than babysit a browser bot. If the vendor has an official API, use that. Supergood's own documentation, read 8 September 2026, says "If a vendor offers an official API, we highly recommend it."

There is a second buyer hiding in the product. We asked the auction integration to page through two provinces, average opening prices and estimated values, and return the percentage difference. Its figures matched our own recalculation to two decimal places. An integration that averages auction prices has stopped being an API and become a small automation with a REST front door, and that is a purchase for a team with a process to automate and nobody free to write it.

What will bite is old software and time. On an ASP.NET WebForms parcel portal, the recorder caught the form POSTs but missed the page loads between them, so the agent had to infer those steps. They worked first time, but sites like that, with no background requests at all, are where screen scraping has always been most brittle. And every repair we watched happened during the build. Whether a live integration gets fixed after a vendor changes something, rather than merely flagged, is the claim that matters most, and we have not yet seen it tested.

Editor's Note: The obvious story about Supergood is code generation, and in our field test it was good: two recorded searches, and the agent pulled a whole portal's category tree out of its JavaScript. What stuck with me came after. When we recently sent one of our deployed integrations the docs' own placeholder service account, all ten failures were logged, classed as user errors, escalated, and annotated with the one sentence you want from triage: retrying will not help. Unofficial integrations have always died quietly in production, so judge Supergood on that after-care, not on the demo. The caveat: we have not yet watched a live integration survive a real change on the vendor's side, so "integrations that maintain themselves" is still a promise. — Rafal Fila, ShadowGen