SerpAPI

Address it

import { create, searchByImage } from "@agntn/web";

const serpapi = await create("serpapi"); // SERPAPI_API_KEY
const results = await serpapi.search("query", { maxResults: 10 });
const in2021 = await serpapi.search("node.js release", {
  startPublishedDate: "2021-01-01",
  endPublishedDate: "2021-12-31",
});
const matches = await searchByImage("https://example.com/image.jpg", { provider: "serpapi" });

The key is a query parameter on serpapi.com.

What it reads

What comes back

Search: publishedDate, image (thumbnail), favicon and metadata.{position, source, displayedLink}. startPublishedDate and endPublishedDate go out as Google's custom date range, the only filter here, and includeDomains and friends come back in ignoredFilters. Paging through start. Image matches carry pageUrl, imageUrl, dimensions, source, position and exactMatch.

Gotchas

• The date range takes whole days in US order, so each bound becomes its UTC day as M/D/YYYY. A lone bound goes out alone, because Google takes cd_min or cd_max by itself and there is no reason to make up the other end. publishedDate is Google's own text, Oct 19, 2021 or 6 days ago, and plenty of results have none.
• The only built-in reverse image provider. Everything under Reverse image search applies.
• 250 searches a month on the free plan as of September 2026. A burst answers 429 fast.
• maxResults is applied on our side to whatever Google or Lens returns. Google pays little attention to num (a maxResults: 1 call still brought back nine organic results) and hands a different page for a start off the ten grid, so a small maxResults walks each Google page in slices: the continuation fetches the same page again, which SerpAPI serves from its cache for an hour without charging, and start moves only once the page is used up.
• No matches is an empty list, not an error. An image Lens cannot fetch looks the same, so [] may also mean the provider never saw the picture.

Where it lives

src/providers/serpapi.ts.