---
title: "Web Search API for AI Agents and LLMs, Priced and Compared"
url: https://busyless.space/seo-apis/web-search
updated: 2026-11-20
source: busyless.space
---

# Web search API for AI agents: page text, MCP and real cost

An agent needs a web search API that returns text the model can read, not only ten links. Exa API, Brave Search API, Perplexity Search API and Tavily API return page text in the call or on the same key; the SERP APIs return Google's results page with snippets and need a scraping step for the rest. At 30,000 searches a month, page text starts at $120 on Exa API, and snippets alone at $30.00 on Serper.

- search providers compared: 15
- return page text, not just snippets: 4
- ship an official MCP server: 11
- host a remote MCP endpoint: 9

## What an AI search API returns that a SERP API does not

Two kinds of product answer the same request here. Exa API, Tavily API, Brave Search API and Perplexity Search API are built for models: each answers from its own index or its own selection of sources, and returns text a model can read, either page text or query-relevant extracts. An AI search API like these is sold per request and does not tell you where a page ranks on Google.

The SERP APIs relay Google's results page: positions, titles, snippets, AI Overviews. That is what you want when the answer has to match what a person sees on Google, or when ranking itself is the data. A snippet is two lines, though, so an agent that must read the page needs a second step.

## Every search provider on this page

15 providers: 4 built for agents and 11 SERP APIs that sell to agent builders. ScrapingBee, Apify, Zyte and Scrapfly also return Google results, but sell scraping first, so they are on the [SERP API page](https://busyless.space/seo-apis/serp) instead.

- [DataForSEO](https://busyless.space/seo-apis/serp/dataforseo) (Markdown: https://busyless.space/seo-apis/serp/dataforseo.md): Twelve pay-as-you-go APIs covering SERPs, keywords, backlinks, on-page and AI search. No subscription, no monthly minimum, and the per-request rates are the lowest of any full-stack provider here. From $0.0006 per SERP.
- [SerpApi](https://busyless.space/seo-apis/serp/serpapi) (Markdown: https://busyless.space/seo-apis/serp/serpapi.md): The most thoroughly documented SERP API, covering Google AI Mode and AI Overviews alongside dozens of engines. It indemnifies US scraping claims up to $2m from the Production plan, and is itself in court with Google. From $25/mo.
- [Serper](https://busyless.space/seo-apis/serp/serper) (Markdown: https://busyless.space/seo-apis/serp/serper.md): Prepaid Google search credits with sub-two-second responses and no subscription. The cheapest credible SERP data here at volume, though the entry rate is $1 per 1,000, not the headline $0.30. From $1.00 per 1,000.
- [Bright Data SERP API](https://busyless.space/seo-apis/serp/bright-data) (Markdown: https://busyless.space/seo-apis/serp/bright-data.md): SERP data on top of the largest commercial proxy network, with unlimited concurrency and parsed JSON for Google and Bing. You pay only for successful requests, and there is no indemnity. From $1.50 per 1,000.
- [Oxylabs SERP Scraper API](https://busyless.space/seo-apis/serp/oxylabs) (Markdown: https://busyless.space/seo-apis/serp/oxylabs.md): Structured SERP parsing on a monthly plan whose fee becomes credit, spent at a per-target rate. Google is $1.00 per 1,000 results on the $15 Starter plan, falling to $0.50 at Enterprise. From $15/mo.
- [ValueSERP](https://busyless.space/seo-apis/serp/valueserp) (Markdown: https://busyless.space/seo-apis/serp/valueserp.md): Monthly credit packs for Google SERPs with rate limits measured per minute rather than per second. Reasonable at volume, and the entry rate is $2.24 per 1,000 on monthly billing, not the advertised $0.50. From $2.24 per 1,000.
- [Scrapingdog](https://busyless.space/seo-apis/scrapingdog) (Markdown: https://busyless.space/seo-apis/scrapingdog.md): A budget SERP API with dedicated endpoints for Google and its AI Overviews and AI Mode, plus Bing, Baidu, Yahoo and DuckDuckGo. Plans start at $40 a month, and only Standard and above carry unused credits over. From $40/mo.
- [ScraperAPI](https://busyless.space/seo-apis/scraperapi) (Markdown: https://busyless.space/seo-apis/scraperapi.md): A scraping API where the destination sets the price: an ordinary page is 1 credit and any Google or Bing request is 25. Generous for page fetching, costly for SERP volume, with structured Google and AI answer endpoints. From $49/mo.
- [SearchApi](https://busyless.space/seo-apis/serp/searchapi) (Markdown: https://busyless.space/seo-apis/serp/searchapi.md): A SERP API that covers ChatGPT Search and Perplexity alongside Google, Bing and a long list of other engines, which makes it one of the more practical ways to monitor AI answer engines through a normal search endpoint. From $40/mo.
- [Scale SERP](https://busyless.space/seo-apis/serp/scale-serp) (Markdown: https://busyless.space/seo-apis/serp/scale-serp.md): ValueSERP's sibling from Traject Data, with more Google search types (Scholar, Autocomplete, Places detail) and a batch endpoint. It costs more per search than ValueSERP at every volume we can compare. From $29/mo.
- [Zenserp](https://busyless.space/seo-apis/serp/zenserp) (Markdown: https://busyless.space/seo-apis/serp/zenserp.md): A straightforward SERP API with a small free tier and simple search-count plans: $49.99 buys 25,000 searches. Nothing unusual, which is occasionally the point. From $49.99/mo.
- [Exa API](https://busyless.space/seo-apis/web-search/exa) (Markdown: https://busyless.space/seo-apis/web-search/exa.md): A search engine for AI with its own index, sold per request in dollars. $4 (lite) or $7 (auto) per 1,000 searches includes page content for ten results, and asking for more results is what raises the bill. From $4 per 1,000 searches.
- [Tavily API](https://busyless.space/seo-apis/web-search/tavily) (Markdown: https://busyless.space/seo-apis/web-search/tavily.md): A web access API for AI agents covering search, extract, map, crawl and research, billed in credits. One credit buys a basic search, five extracted pages or ten mapped URLs. From $0.008 per credit.
- [Brave Search API](https://busyless.space/seo-apis/web-search/brave-search) (Markdown: https://busyless.space/seo-apis/web-search/brave-search.md): Web search from Brave's own index, sold as prepaid credits at $5 per 1,000 requests. The same price buys a results page or LLM Context, which returns extracted page text sized to a token budget. From $5 per 1,000 requests.
- [Perplexity Search API](https://busyless.space/seo-apis/web-search/perplexity) (Markdown: https://busyless.space/seo-apis/web-search/perplexity.md): Ranked web results from Perplexity's index as JSON, each with text extracted from the page for your query. $5 per 1,000 requests, $1 on Fast Search, and one request can carry five queries. From $5 per 1,000 requests.

## What a month of agent searches costs

Set how much work the agent does. Tick the box if the model reads pages rather than snippets.

10,000 agent tasks a month at 3 searches each (30,000 searches). Snippets only: Serper $30.00; Bright Data SERP API $45.00; DataForSEO $60.00; ValueSERP $75.00; Scrapingdog $90.00; Oxylabs SERP Scraper API $100; SearchApi $100; Exa API $120; ScraperAPI $149; Zenserp $150; Brave Search API $150; Perplexity Search API $150; Tavily API $220; Scale SERP $249; SerpApi $275. With 5 pages of text per search: Exa API $120; Brave Search API $150; Perplexity Search API $300; Tavily API $460; the SERP APIs return snippets only from their search call.

For snippets, the SERP APIs are the cheap end: Serper costs $30.00 for 30,000 searches a month. Once the model reads pages, only 4 providers price it in one place, from $120 on Exa API to $460 on Tavily API. Tavily API and Perplexity Search API charge for page text per URL, so the bill grows with every page the agent opens.

Tavily API is priced through its Extract endpoint here. Its search can also return raw page content, and its credit page prices search by depth alone, but the docs never say that content is free, so we did not assume it.

## Whose results, what text, which MCP server

The three questions an agent builder asks first, provider by provider.

| Provider | Results from | Page text | Official MCP server |
| --- | --- | --- | --- |
| [DataForSEO](https://busyless.space/seo-apis/serp/dataforseo) | Google's results page | Separate On-Page endpoint returns Markdown | Yes, hosted at mcp.dataforseo.com |
| [SerpApi](https://busyless.space/seo-apis/serp/serpapi) | Google's results page | Separate Extract API returns Markdown | Yes, hosted at mcp.serpapi.com |
| [Serper](https://busyless.space/seo-apis/serp/serper) | Google's results page | Snippets only | None found |
| [Bright Data SERP API](https://busyless.space/seo-apis/serp/bright-data) | Google's results page | Web Unlocker returns Markdown on the same account | Yes, hosted at mcp.brightdata.com |
| [Oxylabs SERP Scraper API](https://busyless.space/seo-apis/serp/oxylabs) | Google's results page | AI Studio search can return Markdown per result | Yes, hosted at mcp.oxylabs.io |
| [ValueSERP](https://busyless.space/seo-apis/serp/valueserp) | Google's results page | Snippets only | None found |
| [Scrapingdog](https://busyless.space/seo-apis/scrapingdog) | Google's results page | Scraping API returns Markdown on the same key | Yes, run locally |
| [ScraperAPI](https://busyless.space/seo-apis/scraperapi) | Google's results page | Scraper endpoint returns Markdown on the same key | Yes, hosted at mcp.scraperapi.com |
| [SearchApi](https://busyless.space/seo-apis/serp/searchapi) | Google's results page | Scrape API in beta, on request | Yes, hosted at searchapi.io/mcp |
| [Scale SERP](https://busyless.space/seo-apis/serp/scale-serp) | Google's results page | Snippets only | None found |
| [Zenserp](https://busyless.space/seo-apis/serp/zenserp) | Google's results page | Snippets only | None found |
| [Exa API](https://busyless.space/seo-apis/web-search/exa) | Exa's own index | Included at the search price | Yes, hosted at mcp.exa.ai |
| [Tavily API](https://busyless.space/seo-apis/web-search/tavily) | Tavily's own selection of sources | Per URL, $1.60 per 1,000 | Yes, hosted at mcp.tavily.com |
| [Brave Search API](https://busyless.space/seo-apis/web-search/brave-search) | Brave's own index | Included at the search price | Yes, run locally |
| [Perplexity Search API](https://busyless.space/seo-apis/web-search/perplexity) | Perplexity's own index | Per URL, $1.00 per 1,000 | Yes, hosted at api.perplexity.ai/mcp |

_Read from each vendor's docs on 20 November 2026. "Official" means a vendor-owned repository or the vendor's own docs._

The split in the second column matters more than it looks. A SERP API is a search engine API in the strict sense: it reports Google's ranking as it stands. Exa API, Tavily API, Brave Search API and Perplexity Search API answer from their own retrieval, so two of them can return different sources for the same question, and neither will match Google.

## Web search MCP and Google search MCP servers

An MCP server lets an AI client call the search without code. It bills like the API behind it.

Of the 15 providers here, 11 ship an official web search MCP server, and 9 host a remote endpoint you can point a client at with only a key. Scrapingdog and Brave Search API run locally. We found no official server for Serper, ValueSERP, Scale SERP and Zenserp.

People searching for a Google search MCP usually want Google's results inside Claude or Cursor. The servers that do that are run by SERP API vendors, DataForSEO, SerpApi, Bright Data SERP API, Oxylabs SERP Scraper API, Scrapingdog, ScraperAPI and SearchApi among them, each calling its own API. Google's route for its own models is Grounding with Google Search, a Gemini feature rather than an MCP server, and its terms [forbid storing the links](https://busyless.space/seo-apis/google-search-api).

## What the terms let your agent keep

Storing results, caching them or training on them is a license question, not a technical one.

> store, cache, or create a database of Search Results, in whole or in part, other than transient storage required for operation of Customer Applications
>
> Brave Search API Terms of Use (1 September 2026), on what a customer shall not do, https://api-dashboard.search.brave.com/documentation/resources/terms-of-service

[Brave Search API](https://busyless.space/seo-apis/web-search/brave-search)'s self-serve terms also bar using results to train, fine-tune or benchmark models. Brave's FAQ says storage and training rights come with a custom plan, which has no published price.

[Perplexity Search API](https://busyless.space/seo-apis/web-search/perplexity) runs the other way: its Search Addendum lets Perplexity "retain, copy, distribute and otherwise use" your queries and the results, and the zero-data-retention commitment made for its other APIs does not apply to search. Fine for public questions; check it before sending customer data.

Read the terms of whichever provider you shortlist with the same question: may we keep what comes back, and what happens to what we send?

## Free web search API allowances

Enough to build and test an agent; none of them runs one in production.

| Provider | Free searches | How often | Card |
| --- | --- | --- | --- |
| [Bright Data SERP API](https://busyless.space/seo-apis/serp/bright-data) | 5,000 | Every month | Not required |
| [Oxylabs SERP Scraper API](https://busyless.space/seo-apis/serp/oxylabs) | 5,000 | Once | Not required |
| [Serper](https://busyless.space/seo-apis/serp/serper) | 2,500 | Once | Not required |
| [Exa API](https://busyless.space/seo-apis/web-search/exa) | 2,500 | Every month | Not required |
| [Tavily API](https://busyless.space/seo-apis/web-search/tavily) | 1,000 | Every month | Not required |
| [Brave Search API](https://busyless.space/seo-apis/web-search/brave-search) | 1,000 | Every month | Required |
| [SerpApi](https://busyless.space/seo-apis/serp/serpapi) | 250 | Every month | Not required |
| [Scale SERP](https://busyless.space/seo-apis/serp/scale-serp) | 125 | Every month | Not required |
| [SearchApi](https://busyless.space/seo-apis/serp/searchapi) | 100 | Once | Not required |
| [Zenserp](https://busyless.space/seo-apis/serp/zenserp) | 50 | Every month | Not required |
| [ScraperAPI](https://busyless.space/seo-apis/scraperapi) | 40 | Every month | Not required |
| [Scrapingdog](https://busyless.space/seo-apis/scrapingdog) | 10 | Once | Not required |

_Credit allowances converted to basic searches at each vendor's own rate. Read in November 2026._

A free web search API allowance that resets monthly is the useful kind for an agent in development. Exa API, Tavily API and Brave Search API reset every month among the providers built for agents; Perplexity Search API publishes no free tier.

## Choosing a search API for AI agents

Four questions, in the order they rule providers out.

1. **Does the model read pages, or only snippets?** If it reads pages, start with the providers that return page text in the call; it saves a scraping step and a second bill.
2. **Must the answer match Google?** A SERP API is the only way to see what Google ranks. A search API for LLM grounding answers from its own index, which is fine for most questions.
3. **May you keep the results?** Caching, storing and training are licensed separately on some providers. Read the terms before you design the cache.
4. **MCP, SDK or plain HTTP?** A hosted MCP endpoint is the fastest way into Claude, Cursor or an agent framework; the bill is the same as the API.

Many agents use two providers: a web search API for LLM reading, and a search engine API that relays Google to check what people actually see. If you want one web search API for AI work that covers both, test it on both jobs before you commit.

## Where to go from here

- [Cheapest SERP APIs](https://busyless.space/seo-apis/cheapest-serp): 13 providers, 11 of them also on this page.
- [Google News APIs](https://busyless.space/seo-apis/google-news): 12 providers, 11 of them also on this page.
- [Google Search APIs](https://busyless.space/seo-apis/google-search-api): 13 providers, 11 of them also on this page.
- [Content automation](https://busyless.space/services/content-automation-services): Where search APIs feed research and drafting workflows.
- [Our LLM visibility tracker](https://busyless.space/tools/llm-visibility): The other side: what AI answers say about you.
- [Exa vs Tavily](https://busyless.space/compare/exa-vs-tavily): Both hand an agent page text with every search. One charges for each result past ten, the other for the call, and your settings decide which is cheaper.

## Web search APIs, answered

### Which search API for AI agents is cheapest when the agent reads the pages?

At 30,000 searches a month with 5 pages of text each, Exa API costs $120, then Brave Search API $150, Perplexity Search API $300, Tavily API $460. The SERP APIs are cheaper per search but return snippets only from the search call.

### Do Tavily and Exa use Google's search results?

No. Exa says it runs its own web index. Tavily returns its own relevance-ranked selection of sources and does not say which index supplies them. Neither reports Google positions.

### Can I store results from the Brave Search API?

Not on the self-serve plans: the terms allow only transient storage and forbid training on results. Brave says storage rights come with a custom plan, priced by its sales team.

### Is the Perplexity Search API the same as Sonar?

No. The Search API returns ranked results with extracts, billed per request with no token charges. Sonar was Perplexity's answer API; its chat completions became the Agent API, and Perplexity says Sonar support ended on 27 September 2026.

### What is the difference between a SERP API and an AI search API for agents?

A SERP API returns Google's results page as data: positions, snippets and features. An AI search API returns sources chosen by its own retrieval with text a model can read. The first tells you what Google shows; the second gives the model something to read.

---

An independent comparison by busyless. No affiliate links and no paid placement. Prices and terms were read from each vendor's own pages; checked 2026-11-20. HTML version: https://busyless.space/seo-apis/web-search
