AgentRouter
Sign in
AnyAPIAnyAPI

Google Search

/google.search

Run a Google web search and get the organic results (title, link, snippet, position) as clean JSON. Returns about 10 results per call - Google stopped honoring bulk result counts in September 2025, so a limit above 10 is accepted but returns no more than a page. Send page 2, 3, ... to walk further, or use google.search_100 for up to 100 ranked results in one call, which is normally cheaper past roughly 20 results.

Finished calls, last 7 days
0

Cancellations excluded. Execution times cover successful calls.

Use it with your agent

Set up https://www.agentrouter.to/SKILL.md, then use AnyAPI’s /google.search endpoint through AgentRouter.

Inputs

glOptional
Two-letter country code for result localization (e.g. us, gb, de).
hlOptional
Two-letter interface and results language code (e.g. en, es, de).
pageOptional
Which page of about 10 organic results to return, starting at 1. Send page 2, 3, ... to walk deeper; each page is billed as one request and any source can serve any page. Replaces cursor.
limitOptional
Page size to ask for. Google returns about 10 organic results per page and stopped honoring larger counts in September 2025, so a limit above 10 still returns about 10. Some sources return their whole page of about 10 whatever limit says: a page is never trimmed after the call, so nextCursor always continues exactly where this response ended. To go deeper, send page (about 10 results per page, each billed as a request) or use google.search_100, which returns up to 100 ranked results in a single call for one flat charge and is cheaper past roughly 20 results. Price is flat per request.
queryRequired
The Google search query.
cursorOptional
Deprecated: send page instead. A continuation token from an earlier response's nextCursor is still accepted and still continues on the source that issued it; page is ignored when cursor is sent alongside it.
locationOptional
Fine-grained location to localize results to, more specific than the country-level gl. Send it as an Active Canonical Name from Google's geo-target list: place, then each parent region, then country, with no space after each comma, e.g. 'New York,New York,United States', 'Austin,Texas,United States', 'London,England,United Kingdom'. Common variations are translated to that name when they point to exactly one place: spaces after commas, missing accents, a US state abbreviation ('Orlando, FL, United States'), or a skipped region ('Sydney, Australia'). A value that names no place, or more than one, is rejected rather than quietly searched from somewhere else.
timeframeOptional
Restrict results to a recent time window: 1h, 1d, 7d, 1y, or all. Default all (no time restriction).
autocorrectOptional
Toggle Google spelling autocorrect (default true). Set false to search the exact query without correction.
requireCursorOptional
Deprecated. Every source for this search returns a nextCursor, so this changes nothing about the price or which source serves you; it stays accepted so callers that already send it keep working.
preferLatencyUnderMsOptional
Optional; omit it and routing is unchanged, with the cheapest source serving. Prefer sources whose typical response time (median over the trailing 30 days, as published on this endpoint's lane health) is under this many milliseconds; among those, the cheapest serves. This can raise your price: when the cheapest source misses the target, a faster and dearer one serves, and you are quoted and charged its price. If no source is that fast the request is still served, by whichever source offers the best speed for its price - it is never refused for being slow. Sources we have not timed are tried last. This is a preference, not a guarantee: the median describes past requests and is not a ceiling on this one, and it excludes any wait this request itself asks for. On a paginated walk it applies to the first page only: later pages stay with the source that page chose, at the price it was quoted.
sourceOptional
Optional. Source ids to prefer, in order, taken from this endpoint's `lanes[].source.id` in /catalog or /apis. Omit it and the cheapest source serves, with automatic failover. Listed sources are tried first in the order given, then the others, unless `allowFallbacks` is false. A single source with `allowFallbacks` false is served only by that source at its price, quoted and charged exactly, with no failover. The price is that of the dearest source that may serve. An id that does not serve this endpoint is rejected as invalid input with no charge; a listed source that is not serving right now is refused with no charge, so omit `source` to be served by another. On a paginated walk, later pages must include the source that served page one, or omit `source`.
allowFallbacksOptional
Optional, default true. When false, only the sources listed in `source` may serve; the request is refused with no charge if none of them can. When true, the listed sources are tried first and any other source may serve after them, at the normal price.
ignoreSourcesOptional
Optional. Source ids to skip for this request, taken from this endpoint's `lanes[].source.id`. The cheapest remaining source serves and the price is that of the dearest remaining source. An id that does not serve this endpoint, or that is also in `source`, is rejected as invalid input with no charge; skipping every source is rejected the same way.

Example input

{
  "gl": "us",
  "hl": "en",
  "limit": 10,
  "query": "best coffee maker"
}

Output

AnyAPI output: found (boolean), data.results (one page of results) and data.nextCursor for the next page; reason when nothing was found.