Search API for AI applications.
Search the web, news or images and get structured results as JSON or Markdown. Scrape the top results or get a grounded answer in the same call.
- 500 free credits every month
- No card to start
- Failed requests cost nothing
POST /search
{
"query": "stripe payments api"
}
Search results your AI can actually use.
Most search APIs stop at titles, links and snippets. Geekflare fetches the pages behind the results, or get a grounded answer to the query, all in the same request.
Search and scrape in one call
Turn on scrape and get the top results with their page content ready for LLMs. No second round of requests to fetch every link.
scrape: trueGrounded answers
Get an AI answer based on live search results. Good for answer engines, research tools, and chat assistants.
groundedAnswer: trueWeb, news and images
Use the same endpoint to search the web, news, or images depending on what your application needs.
source: web | news | imagesCountry and time filters
Search from a specific country and limit results by time from the last hour to the last year, or a custom window such as the past seven days.
location: inDomain and category filters
Keep results to the sources you trust, exclude sites you don't want, or narrow the search to code, and research.
category: code | researchLLM-ready output
Ads and noise are stripped before results reach you. Take them as JSON or Markdown for a model's context.
format: json | markdownSearch from where your users are.
The same query returns different results in different countries. Set a location and you get the results people there see, for local SEO, market research and regional news.
locationtakes an ISO country code such as"de"or"in"- Pair it with
source: "news"andtimefor regional news - Included in every search at no extra cost
- 0
- Countries
- Country
- Level of targeting
Search parameters you'll use most
See what each search API payload does, what it costs, and when to use it without digging through the API reference.
| Parameter | Type | What it does | Credits |
|---|---|---|---|
| query | string | What to search for. Up to 2,048 characters. | — |
| limit | number | Number of results, from 1 to 100. Defaults to 10. | included |
| source | string | web, news or images. Defaults to web. | included |
| location | string | Country to search from, as an ISO code such as "us" or "de". | included |
| time | string | h, d, w, m or y, or a custom window such as d7 for the last seven days. | included |
| category | string | general, code, pdf, research, linkedin or wiki. | included |
| includeDomains | string[] | Only return results from these domains. excludeDomains does the opposite. | included |
| format | string | json or markdown. Defaults to json. | included |
| scrape | boolean | Also fetch the content of the top results. Set how many with scrapeLimit, from 1 to 10. | +2 |
| groundedAnswer | boolean | Synthesize an AI answer from the search results. | +3 |
What people do with the Search API.
An answer, not a page of links
Search the live web and get a grounded answer synthesized from the results in a single request. Give answer engines and support assistants current information without building the search-to-answer pipeline yourself.
- One call instead of search, fetch and summarize
- Grounded in current web results, not model knowledge alone
- Super fast, results in seconds
POST /search
{
"query": "latest openai model",
"groundedAnswer": true
}Your first search, in a few lines.
# pip install geekflare-api
from geekflare_api.client import GeekflareClient
from geekflare_api.models import SearchRequestDto
with GeekflareClient(api_key="<api-key>") as client:
result = client.search(
SearchRequestDto(
query="best running shoes"
)
)
print(result)One API. Every stack.
The same key works from an agent, a workflow builder, an SDK or your own service. Pick the lane you already live in.
MCP server
One remote endpoint exposes Geekflare's tools to any MCP client. Your agent discovers them itself: no glue code, no per-tool wiring.
Claude
Cursor
Codexany MCP clientLLM & agent pipelines
LLM-ready Markdown straight into a retrieval pipeline, with the boilerplate stripped so you spend context on content, not navigation.
Workflow builders
Run any endpoint as a step in a visual scenario. Useful when the person who needs the data does not write code.
SDKs & REST
First-party typed SDKs for Python and Node. Everything else talks plain REST, from any language.
PythonNode.jsPostman2 credits a search. Here is what that buys.
Credits are shared with every other Geekflare API. Failed requests cost nothing.
Growth covers up to ~50,000 searches a month (100K credits), about $1.38 per 1,000 searches. Or $58/mo billed yearly.
- 500 credits / month
- 1 team member
- 7 days log retention
- 1 request per second
- 10K credits / month
- 3 team members
- 30 days log retention
- 5 requests per second
- 100K credits / month
- 5 team members
- 30 days log retention
- 10 requests per second
- 1M credits / month
- 25 team members
- 90 days log retention
- 25 requests per second
Teams that stopped fixing SERP parsing.
“Its been couple of years we are using Geekflare API, we like its stable production ready performance and cost-effectiveness of accessing multiple APIs through a single plan.”
Jitendra RauloDirector, Aarav Infotech“I've been using Geekflare for both their API tools and their multi-AI chat platform. The URL-to-Markdown API has been a game changer for my content pipeline.”
“Found Geekflare API to get markdown from URL for my AI agents. It is fast and cheaper and works on almost every website.”
Ram DasiArchitect, PA ConsultingBefore you integrate search.
Everything else is in the endpoint docs, and our support team replies to every email.
Open the endpoint docsscrape it costs 4 credits, and a grounded answer costs 5 credits per request. Failed requests cost nothing.scrape and we also visit the top results and return their full page content in the same response, so your AI gets the actual context, not just a list of links.groundedAnswer: true and we use AI to synthesize an answer from the live search results, so you get a direct answer in one request instead of searching, fetching and summarizing yourself.source. Filter by country with location, by recency with time, by type with category (code, pdf, research, linkedin, wiki), and by site with includeDomains and excludeDomains.limit (10 by default). With scrape on, scrapeLimit sets how many of them are fetched in full, from 1 to 10.format: "markdown" the HTML, navigation and ads are gone, which cuts token usage and helps the model focus on the content.scrape on, response time also depends on how fast the result pages load.250 searches a month, free.
The free tier renews every month and needs no card. Give your agent live web results in a single call.