The engine behind the No-Code Tool, exposed as functions you can call. Each one does a whole job: scrape a paginated list, solve a captcha, walk a date picker, run an LLM prompt without leaving the session. Call them from Node, from any CDP client, or hand them to your agent over MCP.
1import { AxiomApi } from 'axiom-api';2 3const axiom = new AxiomApi(process.env.AXIOM_API_KEY);4 5await axiom.browserOpen();6try {7 await axiom.goto("https://example.com/products");8 await axiom.click("button[data-load-more]");9 const rows = await axiom.scrape({ maxResults: 200 });10 console.log(rows.length);11} finally {12 await axiom.browserClose();13}Single calls
Nine of the twenty steps, with the line of code each one takes. The arguments are the selectors and values you already have. The steps that read something hand the data back, the rest act on the page and move on.
The step finds the challenge, gets the answer, and submits it. Your script carries on as though it were never there.
await axiom.solveCaptcha(key);One call loads the page, waits for it, and clicks through the pages until it has your results.
await axiom.scrape(url,
".row", ".next", 500, {});
Classify a page, summarize reviews, or pull a field out of messy text, mid-run, with no model call of your own to stand up.
await axiom.integrateAI({
prompt: "Is this in stock?"
});
The widget that breaks everyone's scraper. Hand it a date and the step clicks its way to the right cell.
await axiom.datePicker(sel, date);Range sliders, kanban cards, map handles, drop zones. Two selectors, real mouse events.
await axiom.clickAndDrag(a, b);For the controls that hide behind a hover state and re-render the moment you touch them.
await axiom.clickEngagementButton(sel);Some pages will only hand over the data through a copy button. This reads what landed there.
await axiom.getClipboardContents();Some pages take Chrome down with them. Get a fresh one inside the same session and keep the run alive.
await axiom.restartBrowser();An isolated cloud Chrome with clean cookies, held open across as many steps as you like.
await axiom.browserOpen();Why steps
Each step does a whole job rather than a single browser instruction, and it runs on the same engine as the No-Code Tool.
In Puppeteer or Playwright you write the mechanics yourself: waitForSelector, $$eval, click next, wait, loop. A step is that sequence already written. Pass it what it needs: click takes the button selector, scrape takes the data selector, the next-page selector, and a result count. A dozen lines come out as one.
The steps run on the same engine as the No-Code Tool, so they wait for elements, retry flaky actions, and time out sensibly by default. Long-running steps poll transparently until they finish. You handle business logic, not browser flakiness.
Steps are ordinary async functions, so they sit in the middle of the JavaScript you already write. Pull rows from your database and loop them through a step. Drop to raw CDP with Puppeteer or Playwright when a step doesn't cover something, then carry on with steps in the same session, on the same browser.
Give your agents hands on a real browser. The steps become tools an MCP client calls on its own: scrape a page for the data it needs, or chain a series to log in, fill a form, and read the result back. Claude, Cursor, or an agent you built yourself, all on your account key.
How to call it
Same steps, same cloud browsers, same key. Pick whichever fits how you already work.
The canonical wrapper. Named methods, HTTP boilerplate handled, transparent retries when a long-running step times out.
Browse every step functionPrefer full DevTools Protocol control? Point any CDP-speaking client at the same cloud browser WebSocket.
Read the Chrome API docsExpose step functions as MCP tools so Claude or another LLM client can drive a session step by step. The axiom.ai MCP server arrives in the next release.
Worked examples
Log in, fill a form, call AI mid-run, or drop the steps into code you already have.
1await axiom.browserOpen();2try {3 await axiom.goto("https://example.com/login");4 await axiom.enterText("#email", "user@example.com");5 await axiom.enterText("#password", process.env.PW);6 await axiom.click("button[type=submit]");7 await axiom.wait(2000);8 9 const rows = await axiom.scrape(null, { hierarchy: ".product-card" }, null, 50, {});10} finally {11 await axiom.browserClose();12}13 1await axiom.goto("https://example.com/apply");2await axiom.enterText("#full-name", row.name);3await axiom.selectList("#country", row.country);4await axiom.click("button[type=submit]");5 1const [[title], [description]] = await axiom.scrapeMetadata([2 { id: "general_metadata_title" },3 { id: "meta_description", value: 'meta[name="description"]' }4]);5const verdict = await axiom.integrateAI({6 aiFunction: "generate",7 prompt: `Product page or category page? ${title} ${description}`8});9 1const targets = await db.query("SELECT url FROM watchlist");2 3await axiom.browserOpen();4try {5 for (const { url } of targets) {6 await axiom.goto(url);7 const [[price]] = await axiom.scrape(null, { hierarchy: ".price" }, null, 1, {});8 if (Number(price) < 100) await notifySlack(url, price);9 }10} finally {11 await axiom.browserClose();12}13 The surface
Five groups. Everything you need to open a browser, get to a page, act on it, take data off it, and close it down.
Open a cloud browser and give it back when you're done.
browserOpen()Start an isolated cloud browserbrowserClose()End the session and stop the meterrestartBrowser()Recover a wedged browser, keep the sessionGet to the page you want to work on.
goto(url)Load a URL and wait for itswitchBrowserTab(selectTab)Move between open tabsEverything you'd do with a mouse and keyboard.
click(select)Click one elementclickMultiple(select)Click every match in turnclickEngagementButton(select)Like, follow, and similar controlshover(select)Trigger hover menus and tooltipsclickAndDrag(start, end)Drag between two pointsenterText(selectTextField, text)Type into a fieldpressKeys(key)Send raw key pressesselectList(select, text)Choose from a dropdowndatePicker(...)Walk a calendar widget to a dateTake data off the page and into your code.
scrape(url, selector, pager, max_results, settings)Load, wait, page through, return recordsscrapeMetadata(metadata)Title, description, and other page metadatagetClipboardContents()Read what the page copiedThe parts that usually need a third-party service.
integrateAI(aiOptions)Run an LLM prompt mid-sessionsolveCaptcha(apiKey)Send the captcha and submit the answerwait(time)Pause for a fixed periodDocs
Quickstarts, worked examples, and full endpoint references for each layer. Start with the quickstart, generate a key, and build from there.
Start with the quickstart