WebSearch
APIs
Submit an API
★ KEIRO20 = 20% off Keirolabs Keirolabs $0.03/1kSerper $0.30/1kZenserp $0.90/1kParallel $1.00/1kPerplexity $1.00/1kSearchApi $1.40/1kSerpApi $1.96/1kExa $4.00/1kBrave $5.00/1kLinkup $5.00/1kYou.com $5.00/1kTavily $5.00/1kOpenAI Web $10/1kClaude Web $10/1k Full price index →

Blog/Explainer

What is a web search API? Types, prices and how to pick one (2026)

A web search API turns a query into results your code can read. Here's how the three kinds work (own index, scraped Google, and AI search), what each costs per 1,000 searches, and how to pick one.

TL;DR

  • A web search API takes a query over HTTP and returns search results as JSON: titles, URLs, snippets, and sometimes full page text or an answer.
  • There are three kinds: own-index APIs (Brave, Exa, Keirolabs), SERP scrapers that return Google's page (Serper, SerpApi), and AI search APIs built for LLMs (Tavily, Parallel, Linkup).
  • Prices run from $0.03 to $25 per 1,000 searches. The cheapest we track is Keirolabs lite at $0.03–0.10.
  • Pick by who reads the results: people, a ranking tool, or a model.
3Kinds of web search API
$0.03Lowest price per 1,000 we track
800×Gap between cheapest and priciest search
0Self-serve whole-web APIs from Google or Bing

A web search API is the plumbing that lets software search the web the way you do in a browser. Your code sends a query, the API sends back results as JSON, and you do whatever you like with them: show them to users, feed them to a model, track rankings, or build a dataset.

That’s the whole idea. The interesting part is where the results come from, because that decides the price, the speed, the legal risk and how good the results are for your job.

How a web search API works

Every web search API does the same three things:

  1. Takes a query over HTTP, usually with filters like country, language, freshness or a domain list.
  2. Finds matching pages, either in its own index or by asking a search engine.
  3. Returns structured results: title, URL and snippet at minimum. Many also return full page text, extracted highlights, or a written answer with citations.

A request looks like this on almost every provider:

curl -X POST https://api.keirolabs.cloud/api/v2/search/fast \
  -H "Authorization: Bearer $KEIRO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "latest rust release"}'

What changes between providers is step 2.

The three kinds of web search API

KindWhere results come fromExamplesPer 1,000
Own indexThe provider crawls the web and runs its own search engineBrave, Exa, Keirolabs$0.03–12
SERP scraperThe provider fetches Google’s (or Bing’s) results page and parses itSerper, SerpApi, DataForSEO$0.30–25
AI searchBuilt for LLMs: results plus cleaned page text, ranked excerpts or an answerTavily, Parallel, Linkup, You.com$1–8

The lines blur. Exa and Keirolabs run their own index and return page text for LLMs. Keirolabs also has a Google SERP endpoint on the same key. But the three buckets are still the best way to think about it.

Own-index APIs

The provider runs a crawler, stores billions of pages and ranks them itself. Nobody upstream can switch it off, which matters a lot after Bing’s API was retired and Google’s Custom Search closed.

  • Pros: no scraping risk, stable pricing, often the cheapest per call, fast.
  • Cons: results won’t match Google’s exactly.

SERP scrapers

These give you Google’s results page as JSON: organic results, ads, People Also Ask, maps, shopping. They’re the tool for SEO, rank tracking and anything where “what Google shows” is the point. More in what is a SERP API.

  • Pros: exactly what Google shows, rich result types.
  • Cons: you depend on someone scraping Google, which Google doesn’t permit; snippets only, no page text.

AI search APIs

Built for the case where a model reads the results. They return clean page content, compressed excerpts or a cited answer, so you can skip running a scraper. Most of them run on their own index or an aggregation of others.

  • Pros: less glue code, fewer tokens wasted, built-in answers.
  • Cons: pricier per call than raw search; quality claims are mostly self-reported (see our benchmarks).

What a web search API costs

Prices are list prices from each vendor’s own pricing page, checked November 5, 2026.

APICheapest mode per 1,000Free allowance
Keirolabs$0.03–0.10 (lite)750 credits
Serper$0.30–1.002,500 queries once
Parallel$1 (turbo/fast)Free monthly requests
Exa$4 (lite)$10 a month
Brave$5$5 of credits a month
Tavily$5–81,000 credits a month
OpenAI / Claude web search$10 + tokensnone

The search price is rarely your biggest cost. If a model reads 3,000 tokens of results per search at $3 per million input tokens, that’s $9 per 1,000 searches just to read them. An API that returns tight, relevant text can beat a cheaper one that dumps whole pages. The calculator prices every mode at your volume.

Which kind do you need?

Answer one question: who reads the results?

People read them

Own index (Brave)

Results pages, site search, a research tool for humans. Independent, stable, no scraping risk.

A ranking tool reads them

SERP scraper (Serper)

Rank tracking and SEO need Google’s exact page. Nothing else will do.

A model reads them

AI search (Exa, Parallel, Tavily)

Page text or excerpts in one call, so your agent doesn’t need a scraper.

Price decides it

Keirolabs

Own index from $0.03 per 1,000, plus page text, answers, deep search and Google SERP on one key.

Where Keirolabs fits

Keirolabs is Keirolabs’ search, browser and data API. It’s worth knowing about whichever kind you need, because it covers all three on one key and one credit balance:

  • Own index: lite search at $0.10 per 1,000 on Essential, $0.06 on Pro and $0.03 on Startup. Fast search is 0.5 credits.
  • AI search: search + page markdown (1.5 credits), cited answers (1 credit), deep search with ranked evidence (2 credits).
  • SERP: Google results as JSON (2 credits).
  • Extract and MCP: any URL to markdown with JS rendering when search isn’t enough, plus an MCP server for Claude Code, Cursor, Codex and LangChain.

Its deep search found the answer for 96.5% of 200 SimpleQA questions in a public benchmark repo, the highest score we track with code and results you can check.

750 free credits at signup, no card. Through Black Friday, code KEIRO20 takes 20% off any paid plan: Keirolabs pricing.

FAQ

Is there a free web search API?

Not for production, but several have useful free allowances. See free web search APIs for every option and how far each one goes.

Does Google have a web search API?

Not a self-serve one any more. The Custom Search JSON API is closed to new customers and ends January 1, 2027. For Google’s results, use a SERP API.

What’s the difference between a search API and a scraping API?

A search API finds pages for a query. A scraping API fetches a page you already have the URL for. Many projects need both, and some APIs do both. Full breakdown: web scraping API vs search API.

Which web search API is best?

It depends on the job. Our best web search APIs guide picks a winner per use case.

Topic hub

More on search apis for ai and agents

  1. Best web search APIs in 2026, ranked on two benchmarksStart here
  2. Web search in LM Studio and Open WebUI: free and paid setups, step by step
  3. How to search the web from Python: ddgs, googlesearch and 3 APIs (with code)
  4. Tavily MCP, Exa MCP and Keirolabs MCP: setup for Claude Code, Cursor and LM Studio
  5. Tavily vs Perplexity vs Exa: search API pricing, accuracy and which to pick
  6. Keirolabs: pricing, endpoints, benchmarks and how to start
  7. OpenAI web search API: pricing, limits and when a search API is cheaper
  8. Exa vs Tavily: pricing, index, accuracy and which to pick
  9. Exa API pricing and API key: every endpoint, explained
  10. Tavily API key: get one free in 2 minutes, plus the real cost per search
  11. Web search APIs for AI agents: MCP servers, token costs and what actually matters
  12. Free web search APIs in 2026: every free tier, and how far it really goes
  13. Perplexity API pricing (Nov 2026): $5 per 1,000 searches, or $1 on Fast
  14. Tavily alternatives: 8 search APIs for AI agents, compared

Keep reading