Overview
SerpKite is a search API designed from the ground up for AI agents, assistants and retrieval pipelines that need fresh Google results. Most agents eventually hit the same wall: the model knows nothing about yesterday, so you bolt on a search tool, and the existing SERP APIs return enormous, HTML-shaped JSON that burns tokens, bills you for failed requests and expires the credits you already paid for. SerpKite takes the opposite approach. Results come back in a compact, predictable envelope, can be projected down to just the fields you ask for, and can be returned as Markdown that drops straight into a prompt. The same key and the same /v1 API cover twelve Google verticals, page extraction, crawling and monitoring, so a single integration replaces a handful of scrapers and vendors.
Key Features
Twelve Google verticals under one API. Web search, news, images, videos, maps, places, reviews, shopping, scholar, patents, autocomplete and webpage extraction all live under /v1 with the same authentication, pagination, localisation and output options. Agents can pick the right vertical per question without juggling separate providers, keys or response shapes. Localisation parameters such as country, language and location are supported across the verticals that make sense for them, which matters when an agent has to answer for a specific market.
LLM-ready output. Every endpoint can return full JSON, compact JSON or Markdown. Compact JSON removes the noise that wastes context windows, field projection lets you keep only titles, links and snippets, and a token estimate header tells your agent how large the payload is before it is added to a prompt. The include_content option goes one step further and returns the top results as clean Markdown in the same call, so a research agent can search and read in a single round trip instead of fetching each page separately.
Extract, map and crawl. The webpage endpoint converts any public URL into clean Markdown. Site endpoints map the URLs of a domain, extract multiple pages, and run asynchronous crawls that reserve credits up front and refund whatever was not used. Scheduled monitors rerun a search, a news query or a page fetch on a schedule and notify you through signed webhooks when something changes, which is useful for brand monitoring, competitor tracking and keeping an agent's knowledge base fresh.
Remote MCP server. SerpKite exposes a Streamable HTTP Model Context Protocol endpoint with fourteen tools, so Claude, Cursor, VS Code and any other MCP client can search the web, read pages and crawl sites without writing glue code. Add the endpoint and your API key to the client configuration and the tools appear immediately. The same capabilities are available through the REST API for custom agents and workflows.
SDKs and integrations. Official SDKs are available for TypeScript, Python and Go, each with typed responses generated from the same OpenAPI contract that drives the server. There are native integrations for LangChain, CrewAI and the Vercel AI SDK, plus nodes and tools for n8n and Dify, so you can drop SerpKite into the framework you already use. The Python package supports both synchronous and asynchronous clients, and the TypeScript client has no runtime dependencies.
Pricing & Plans
SerpKite is freemium and uses prepaid credits rather than subscriptions. New accounts get 1,000 free credits to start, 1,500 more for linking GitHub or Google, and 1,000 free credits every month after that. When you need more, credit packs start at 10 dollars for 10,000 credits and the price per thousand credits falls to 0.60 dollars at larger volumes. Credits never expire, so you are never forced to use them before a deadline. Failed, empty and blocked searches are not billed, spend caps and per-key monthly limits protect your budget, and low-balance alerts and optional auto-rechar
