Pull live results from 165 engines, including Google, Bing, YouTube, and Amazon, as JSON or Markdown through one REST API or a hosted MCP server.
Visit SerpApiFits SEO teams and AI agent builders who want live Google data with legal cover and accept a higher price per search.
250 free searches a month, then $25 for 1,000 with only successful calls billed. Lower tiers cost $15 to $25 per 1,000 and skip the $2M Legal Shield.
Free plan: 250 searches/mo. Starter $25/mo for 1,000, Developer $75 for 5,000, Production $150 for 15,000. Ludicrous Speed costs 2x. Monthly billing, cancel anytime. 7-day refund if under 20% used.
SerpApi is a search results API. You send a query and an engine name, and it returns what a real user would see as JSON, raw HTML, or Markdown. The Austin company has run since 2017 and handles the proxies, browsers, and CAPTCHA solving for you.
It targets SEO platforms, price trackers, and AI agents that need live search data. Its public engine index listed 165 engines when we checked on November 21, 2026.
location parameter.We checked the public docs, the pricing page, and the open MCP endpoint. The playground asked our automated browser for a CAPTCHA and a sign-in, so we ran no live searches.
Every search goes to the same /search endpoint with an engine parameter. Google alone has more than 60 engines, from Maps and Shopping to Flights, Scholar, and Trends. The rest covers Bing, Baidu, Yandex, Naver, DuckDuckGo, Brave, Amazon, Walmart, YouTube, and Zillow.
Each request runs in a full browser, and SerpApi solves the CAPTCHAs. Results come back in the same request by default, or you can submit searches async and fetch them later.

Some AI Overviews arrive inside a regular Google Search response. Others need a second request to the AI Overview engine with a page_token. That token expires within 1 minute, so your code must use it right away.

Add output=md to any request and SerpApi returns Markdown tables in place of nested JSON. SerpApi's own table shows a Google Search response dropping from 150K to 39K tokens, a 74% cut. The JSON Restrictor parameter trims responses further by keeping only the fields you name.

The MCP server lives at mcp.serpapi.com and takes your API key as a Bearer header. We listed its tools without a key and found four: search, extract, search_table, and search_dashboard. The extract tool scrapes, maps, or crawls a URL, with crawls capped at 100 pages.
Searches default to the google_light engine unless you set another. You can also self-host the server from its GitHub repo or install a Claude Desktop bundle.

SerpApi takes on legal liability for scraping and parsing public search data, with up to $2 million in coverage. It is included from the $150 Production plan up. Free, Starter, and Developer accounts get none.
Coverage applies only to claims under US law in a US court. It also covers collection only, not what you do with the data afterward.

The security page lists SOC 2 Type II, SOC 3, ISO 27001, and ISO 27701 as acquired. GDPR and ISO 42001 are both marked In Progress. Search data expires after 31 days by default.
ZeroTrace Mode stores no queries or results at all. It only appears on the Cloud 1M plan ($3,750 a month) and above.

The free plan gives 250 searches a month at 50 per hour. Paid plans run month to month:
Only successful searches count, and cached results are free for 1 hour. There is no per-search overage. When searches run out, you renew the whole plan early.
Ludicrous Speed doubles the plan price, and Ludicrous Speed Max quadruples it. Refunds are open for 7 days if you used under 20% of your searches.
Pick SerpApi if you need many search engines behind one key, AI Overview data, and legal cover for a production product. The MCP server and Markdown output also make it an easy fit for AI agents. Skip it if you only need bulk Google rankings on a budget, because $10 to $25 per 1,000 searches adds up fast.
Software engineer and founder (15+ years building web apps). Every listing is verified against live product checks, official pricing pages, and real usage limits before publication.
Updated today