SerpKite documentation
Search APIs, cited AI answers, page and PDF extraction, crawls, deep research and monitors for AI agents. Quickstart, endpoint reference, guides and MCP.
SerpKite is a Google search API built for AI agents: clean JSON or Markdown, official TypeScript/Python/Go SDKs, native LangChain, CrewAI and MCP, credits that never expire.
You send a query, and you get back the live Google results page as structured JSON, as token-lean compact JSON, or as Markdown that you can hand straight to an LLM. Search verticals share the same envelope: request (what ran), results (the main list for each search vertical), endpoint extras such as knowledge_graph and people_also_ask, and meta (cost, cache, request ID). Every link is the resolved destination URL.
Beyond search, read HTML pages and PDFs with webpage or extract, discover site URLs with map, ingest a site section with crawl, and schedule recurring checks with monitors. When you want text written for you, Claude adds a cited answer to a search, turns pages into JSON for your schema with extract, or writes a cited report with research. To see what public AI assistants tell people, ask ChatGPT, Gemini, Copilot Search, Google AI Mode or Brave’s AI answer a prompt and get the answer, its sources and mentions of your brand (see AI assistants). These APIs have their own response shapes, described in their endpoint references. Google is the default search provider; opt into other providers when your workflow allows fallback or consensus.
There is also a Google Custom Search JSON API–compatible endpoint for teams moving off CSE before it shuts down on January 1, 2027.
Make your first request
curl https://api.serpkite.com/v1/search \
-H "Authorization: Bearer $SERPKITE_API_KEY" \
-H "Content-Type: application/json" \
-d '{"q":"best espresso machine 2026","country":"us","language":"en"}'Install an SDK with npm install serpkite, pip install serpkite or go get github.com/serpkite/serpkite-go, or call the HTTP API directly. See SDKs.
New accounts get 1,000 free credits, plus 1,500 more when you link GitHub or Google, and 1,000 every month from the next month on. No card required. Get a key, then follow the quickstart.
Start here
Endpoints
Search verticals and webpage accept a POST JSON object or a GET with query parameters under https://api.serpkite.com/v1. Rank, answer, extract, map, crawl and research creation, AI assistant prompts, monitor creation and batch creation use POST; task polling and account/status reads use GET, monitor updates use PATCH, and cancellation/deletion use DELETE. The CSE-compatible endpoint is GET /customsearch/v1. Credits are per results page of 10 unless noted. Unknown parameters are rejected with a 400 that names the right one, so typos never fall back to defaults silently (see Strict validation).
| Endpoint | Method and path | Credits | Returns |
|---|---|---|---|
| Google Search | POST /v1/search | 1 | request, results, answer_box, ai_overview, knowledge_graph, people_also_ask, related_searches, top_stories, places, ads, meta |
| Google News | POST /v1/news | 1 | request, results, meta |
| Google Images | POST /v1/images | 1 | request, results, meta |
| Google Videos | POST /v1/videos | 1 | request, results, meta |
| Google Maps | POST /v1/maps | 1 | request, results, meta |
| Google Places | POST /v1/places | 1 | request, results, meta |
| Google Place Details | POST /v1/place | 1 | request, results, meta |
| Google Reviews | POST /v1/reviews | 1 | request, results, next_page_token, meta |
| Google Shopping | POST /v1/shopping | 1 | request, results, meta |
| Google Scholar | POST /v1/scholar | 1 | request, results, meta |
| Google Patents | POST /v1/patents | 1 | request, results, meta |
| Google Autocomplete | POST /v1/autocomplete | 0.5 | request, results, meta |
| Webpage to Markdown | POST /v1/webpage | 1 | request, url, status_code, markdown, text, metadata, meta |
| Custom Search (CSE-compatible) | GET /customsearch/v1 | 1 | kind, searchInformation, items, queries |
| Answer | POST /v1/answer | 3–10 | answer, citations, results |
| Extract | POST /v1/extract | 1 | results |
| Map | POST /v1/map | 1 | results |
| Crawl | POST /v1/crawl | 1 | a task (202, poll or webhook) |
| Research | POST /v1/research | 30–200 | a task (202, poll or webhook) |
| Monitors | POST /v1/monitors | 1 | results |
| ChatGPT | POST /v1/chatgpt | 4 | text, markdown, sources, search_queries, products, mentions |
| Gemini | POST /v1/gemini | 3 | text, markdown, sources, search_queries, products, mentions |
| Copilot Search | POST /v1/copilot | 3 | text, markdown, sources, search_queries, products, mentions |
| Google AI Mode | POST /v1/ai_mode | 2 | text, markdown, sources, search_queries, products, mentions |
| Brave AI | POST /v1/brave_ai | 2 | text, markdown, sources, search_queries, products, mentions |
To run many queries at half price, queue them with POST /v1/batches (AI assistant answers keep their full price there). Two free endpoints help with bookkeeping: GET /v1/account returns your balance and this month’s usage, and GET /v1/batches/{id} returns the status and result of a batch job.
Build with it
Moving off Google Custom Search
Point https://www.googleapis.com/customsearch/v1 at https://api.serpkite.com/customsearch/v1 and keep your existing CSE client. See Migrate from Google CSE.
Machine-readable docs
Every page here has a Markdown twin: add .md to the URL (for example /docs/quickstart.md), or use the Copy page button above. The whole documentation is also available as llms.txt and llms-full.txt, and the API contract as an OpenAPI 3.1 spec.
Last updated: 2026-11-05