Vercel AI SDK

install @megalania/ai-sdk, then pass megalaniaTools() as tools to generateText or streamText. The key comes from MEG_API_KEY. Blocked pages and errors are free.

Install

Node 22.18 or newer, AI SDK 5 or newer.

npm install @megalania/ai-sdk megalania ai

Use it

import { generateText, stepCountIs } from "ai";
import { megalaniaTools } from "@megalania/ai-sdk";

const { text } = await generateText({
  model: "openai/gpt-5",
  tools: megalaniaTools(),
  stopWhen: stepCountIs(5),
  prompt: "What can the API at https://docs.megalania.io/ do?",
});
console.log(text);

The five tools are scrape, extract, map_site, create_batch and get_batch. extract reads fields by CSS rules or by a JSON Schema. Their inputs use the same names as the API. You can run one yourself, with no model:

import { megalaniaTools } from "@megalania/ai-sdk";

const page = await megalaniaTools().scrape.execute({ url: "https://docs.megalania.io/" }, { toolCallId: "1", messages: [] });
console.log(page.credits_charged, page.text.slice(0, 200));

scrape returns { text, credits_charged, request_id }. get_batch returns the batch with the content of its finished pages.

What a call may cost

Each page costs at most 128 credits by default. That covers every route, and a residential Chrome page up to about 1.5 MiB. A bigger page comes back cut, with a note that says so. Pass megalaniaTools({ maxCredits }) (up to 32767) to allow more. maxCredits is per page, so a batch can cost up to maxCredits times its pages: one create_batch call takes at most 100 pages by default. More URLs are refused, and a crawl stops at 100 pages. Pass megalaniaTools({ maxPages }) to allow more. The model can raise neither. maxChars (default 100,000) cuts what a tool returns, so one page cannot fill the context.

The code is MIT licensed at https://github.com/megalania/megalania-typescript.