Google Custom Search

Give AI agents live web lookup through Google's Custom Search JSON API. Agents run web, image, and site restricted searches against your Programmable Search Engine and return ranked titles, links, and snippets to ground downstream steps.

3 actions API key available

Google Custom Search website ↗ Platform Documentation ↗ Capability data verified 2026-07-28

The Monday competitive sweep starts on schedule, carrying the watch list of tracked terms and competitor domains

Search runs each tracked term and returns ranked titles, links, and snippets with total-result counts

Site Search runs restricted to each competitor's domain, surfacing new pages that never rank for brand terms

The agent diffs this week's links against the standing log to isolate what actually changed

The marketer reviews the flagged findings and approves the deeper pull, because every additional page of results draws on a metered query quota

The approved threads get paginated through Start Index and passed to the scraping step for full-page content

The digest posts to the team channel with new pages, moved results, and the snippets that explain why they matter

What This Integration Enables

An AI agent without live lookup is arguing from memory. Google Custom Search is how FlowRunner agents check their work: a query goes out through your Programmable Search Engine and ranked results come back as structured data, title, link, snippet, with pagination cursors and result totals attached. Because the engine configuration is yours, you decide the universe the agent searches: the whole web, a curated set of sources, or a single domain under watch. - Ground an AI agent's answers in current, attributable search results instead of training-data recall - Run standing competitive sweeps restricted to specific domains, and diff the results over time - Feed top links into scraping and summarization steps for research and monitoring flows - Collect images by query, filtered by size and type, for content and catalog pipelines - Keep query spend visible, with [human-in-the-loop](/concepts/human-in-the-loop) approval before a sweep expands

Without FlowRunner

Research by tab-hopping Competitive and market questions get answered by whoever has time to open twelve tabs, and the findings live in that person's head

Monitoring by memory Someone checks the competitor's site when they remember to, so launches and pricing changes surface weeks late

Agents answering from training data An AI agent asked about the current market reasons from whatever it learned before its cutoff, with no way to check

With FlowRunner

Ranked results inside the flow Every tracked term returns titles, links, and snippets as structured data any downstream step can consume

Standing sweeps with a diff Site-restricted searches run on schedule, and only what changed since last week reaches a human

Every claim carries a link Agent conclusions ground in live results, so the digest cites the page, not the model's recollection

Use Case Scenarios

The research agent that cites its sources

An analyst asks the agent a market question. The agent runs Search on the key terms, takes the top-ranked links, and passes each one to [Firecrawl](/integrations/firecrawl) to extract clean page content. The summary it returns quotes the pages and links to them, so the analyst can verify any claim in one click. The difference between this and asking a chatbot is the difference between an answer and an answer with receipts.

The competitor's site, watched by a robot with a ledger

Every week, Site Search runs with the filter set to Include against each competitor domain on the watch list, one query per tracked topic. New links get appended to [Google Sheets](/integrations/google-sheets) with the date first seen, and the diff against last week posts to [Slack](/integrations/slack). When the competitor ships a pricing page revision or a new product page, the team reads about it in Monday's digest, not in a lost deal three weeks later.

Product imagery sourced with filters, reviewed by people

The content team needs reference imagery for a catalog refresh. The agent runs Search Images for each product term with Image Size set to Large and Image Type set to Photo, collects the image URLs, thumbnails, and source pages, and posts the top matches to the team channel. A person picks the winners and checks the source pages for usage rights. The agent did the gathering; the judgment call about what represents the brand stayed human.

Human-in-Loop Highlight

The costly operation here is the query itself. The Custom Search JSON API is metered: each request returns at most 10 results and draws against a daily quota, with a billed rate once the free allowance is spent. Pagination multiplies quietly. Pulling 100 results for one term is 10 requests; a 50-term sweep paginated five levels deep is 250. An agent left to "be thorough" will be thorough with your quota. So FlowRunner treats sweep expansion as a decision, not a default: the agent runs the budgeted first pass, shows the marketer what changed and what it wants to dig into, and the deeper paginated pull runs only on approval. The spend is small per query and real in aggregate, which is exactly the kind of spend that deserves a witness.

Agent processes routinely

Detects exception requiring judgment

Clear match Continues automatically

Ambiguous Routes to human via preferred channel

Human decides

Agent resumes with decision

Agent Capabilities

3 actions

Web and Site Search

Search Performs a web search against your Google Programmable Search Engine and returns ranked results with title, link, and snippet, plus search metadata and pagination cursors. Results per page are capped at 10 by Google; page with Start Index, keeping start plus num within 100. Supports language filtering and SafeSearch. What comes back depends on how your Search Engine (cx) is configured.
Site Search Runs a web search restricted to, or excluding, a specific site or domain using Google's siteSearch and siteSearchFilter parameters. Include returns only results from the given site; Exclude omits them. Returns the same ranked envelope as Search. The workhorse behind domain monitoring sweeps.

Image Search

Search Images Performs an image search against your Programmable Search Engine and returns image results with the image URL, thumbnail, dimensions, and the page each image was found on. Narrow results by image size and image type; page through with Start Index. Your Search Engine must have image search enabled.

Frequently Asked Questions

What can FlowRunner do with Google Custom Search?

FlowRunner agents can run Search, Search Images, and Site Search in Google Custom Search.

Does connecting Google Custom Search to FlowRunner require OAuth?

No. Google Custom Search connects to FlowRunner with an API key, no OAuth flow required.

Can Google Custom Search trigger a FlowRunner workflow automatically?

Google Custom Search doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.

Start building with Google Custom Search

$100 in credits. No card required. Connect in minutes.

Start Building Free Talk to us