Scraping, voice, maps, browser agent

Firecrawl + Jina Reader, OpenAI Images + ElevenLabs + Whisper + Deepgram, Nominatim + OpenWeatherMap + CoinGecko, and a BYO-Playwright browser agent.

All under @agentskit/tools/integrations. Same pattern as S21 — focused defineTool factories + a bundle helper per provider.

Scraping + parsing

import { firecrawl, reader, documentParsers } from '@agentskit/tools/integrations'

const tools = [
  ...firecrawl({ apiKey: process.env.FIRECRAWL_API_KEY! }),
  ...reader({ apiKey: process.env.JINA_TOKEN }),
  ...documentParsers({
    parsePdf: async bytes => {
      const { default: pdf } = await import('pdf-parse')
      const r = await pdf(Buffer.from(bytes))
      return { text: r.text, pages: r.numpages }
    },
  }),
]

firecrawl_scrape / firecrawl_crawl — managed scraper with JS rendering.
reader_fetch — zero-dep Jina Reader wrapper; returns LLM-ready text.
parse_pdf / parse_docx / parse_xlsx — BYO parser functions so you pick the native-dep story.

Image + voice

import { openaiImages, elevenlabs, whisper, deepgram } from '@agentskit/tools/integrations'

const tools = [
  ...openaiImages({ apiKey: process.env.OPENAI_API_KEY!, model: 'gpt-image-1' }),
  ...elevenlabs({ apiKey: process.env.ELEVENLABS_API_KEY! }),
  ...whisper({ apiKey: process.env.OPENAI_API_KEY! }),
  ...deepgram({ apiKey: process.env.DEEPGRAM_API_KEY! }),
]

openai_image_generate — text → image, returns URL or base64.
elevenlabs_tts — text → MPEG audio bytes (base64 in the result).
whisper_transcribe / deepgram_transcribe — audio URL → transcript.

Binary outputs are base64-encoded so they pass safely through JSON tool results. Persist or stream them on the caller side.

Maps / weather / finance

import { maps, weather, coingecko } from '@agentskit/tools/integrations'

const tools = [
  ...maps({ userAgent: 'myapp/1.0 (contact@example.com)' }),
  ...weather({ apiKey: process.env.OPENWEATHERMAP_KEY! }),
  ...coingecko(), // works anonymously; add apiKey for pro
]

maps_geocode / maps_reverse_geocode — OpenStreetMap Nominatim, free with a required user agent identifying your app.
weather_current — OpenWeatherMap current conditions.
coingecko_price / coingecko_market_chart — crypto prices + series.

Browser agent (BYO Playwright)

Bundling Playwright is a ~200 MB footgun. Instead, the browser agent takes a BrowserPage contract with six methods. Wire your own Playwright/Puppeteer/Chromium DevTools Protocol page into it.

import { chromium } from 'playwright'
import { browserAgent, type BrowserPage } from '@agentskit/tools/integrations'

const browser = await chromium.launch()
const raw = await browser.newPage()

const page: BrowserPage = {
  goto: async url => { await raw.goto(url) },
  click: async selector => { await raw.click(selector) },
  fill: async (selector, value) => { await raw.fill(selector, value) },
  textContent: async selector => (await raw.textContent(selector)) ?? '',
  screenshot: async () => (await raw.screenshot({ type: 'png' })).toString('base64'),
  waitForSelector: async (selector, options) => { await raw.waitForSelector(selector, { timeout: options?.timeoutMs }) },
}

const tools = browserAgent({ page })

Tools: browser_goto, browser_click, browser_fill, browser_read, browser_wait_for, browser_screenshot.

Pair with mandatory sandbox to keep every browser call behind your policy.

Scraping, voice, maps, browser agent

Scraping + parsing

Image + voice

Maps / weather / finance

Browser agent (BYO Playwright)

See also

On this page