Automate any website,even behind a login.

[How it works]

Describe the task.Kaabist runs it.

1. 01

   Describe the task

   Tell Claude Code, Codex or Cursor what you need, in plain words: “download last month’s invoices from our supplier portal on the 1st of every month”.

2. 02

   Your agent builds a program

   It works through the site in a real Kaabist browser and saves the steps as a program: a short JSON file you can read, test and version, not a pile of selectors.

3. 03

   Kaabist runs it

   Call it from your code, the CLI or a schedule. It runs in Kaabist’s browsers, signs in with the details you pass, and sends the results to your app as typed data or a signed webhook.

4. 04

   It keeps working

   If the site moves a button, Kaabist finds it again and sends you the fix to approve. If it asks for a 2FA code, you get a link. If a payment step crashes, it is never pressed twice.

[Where Kaabist fits]

Your agent writes the program.Kaabist keeps it running.

[Why it doesn’t break]

Built for flows,not just pages.

1. 01

   Resumes where it stopped

   When a site asks for a 2FA code, a captcha or an approval, the run pauses and Kaabist emails you a link. You answer, and it continues from its last saved step, not from the start.

2. › claude mcp add kaab …

   goto /login

   fill #email

   click Sign in

   ✓ saved prg_4Rb · v1

   02

   Built by your coding agent

   Connect Claude Code, Codex or Cursor with one line. Your agent tries the site in a real Kaabist browser, and what works is saved as a program you can review and version.

3. 03

   Never submits twice

   Steps that submit, pay or order are marked. If a server crashes in the middle of one, Kaabist does not click again; it pauses and asks you to check what happened.

[Built by your coding agent]

One line in your agent.A program in your account.

$ claude mcp add --transport http kaab https://mcp.kaab.ist/mcp
Added HTTP MCP server kaab
$ claude "renew my subscription on portal.example.test every month"
› browser_open https://portal.example.test/login
› browser_fill #email, #password
› await_input otp                    a person answers from a link
› browser_click Renew                marked side_effect: true
› test_program dry_run               stops before the first side effect
✓ program prg_4Rb · v1 saved · schedule: monthly

[Never submits twice]

A worker crashes on “Pay now”.Kaabist doesn’t press it again.

1. fill

   card details

2. checkpoint

   before_payment

3. fence

   about_to_execute

4. crash

   worker lost

5. park

   side_effect_ambiguous

6. confirm

   you check the order

await page.goto('https://portal.example.test/login');
await page.fill('#root > div:nth-child(2) input', email);
await page.fill('.css-1x9f2k input[type=password]', pw);
await page.click('button.btn-primary');
// 2FA? someone has to be at the keyboard
const code = await waitForSlackMessage();
await page.fill('#otp', code);
await page.click('text=Renew');
// crashed here? the retry renews twice

{ "goto": "https://portal.example.test/login" },
{ "fill": "#email", "value": "{{inputs.email}}" },
{ "fill": "#password", "value": "{{inputs.password}}" },
{ "click": { "role": "button", "name": "Sign in" } },
{ "await_input": "otp" },
{ "fill": "#otp", "value": "{{vars.otp}}" },
{ "checkpoint": "after_login" },
{ "click": { "role": "button", "name": "Renew" }, "side_effect": true }

[Around every program]

The promises rest onthe rest of the platform.

• Self-healing locators

  When a site changes, Kaabist finds the moved element again and sends you the fix to approve.

• API discovery

  A browser run spots the JSON API behind the page. Later runs call it directly, without a browser: faster and cheaper.

• Sessions

  Log in once. Later runs reuse the saved session and skip the login entirely.

• Schedules and change detection

  Run on a schedule, and get an event only when the extracted data actually changed.

• Signed webhooks

  Every result reaches your server as a signed event, retried with backoff until it lands.

• Answer from a link

  When a run needs you, Kaabist emails you a link. Open it, type the code, and the run carries on. No account needed.

[Start in one call]

Any URL in.Structured data out.

import { Kaab } from "@kaabist/sdk";

const kaab = new Kaab(); // KAAB_API_KEY
const job = await kaab.jobs.create({
  url: "https://shop.example.test/p/1",
  extract: { title: "h1", price: { selector: ".price", type: "money" } },
  wait: 30,
});
console.log(job.data);

from kaabist import Kaab

with Kaab() as kaab:  # KAAB_API_KEY
    job = kaab.jobs.create(
        url="https://shop.example.test/p/1",
        extract={"title": "h1", "price": {"selector": ".price", "type": "money"}},
        wait=30,
    )
    print(job.data)

c := kaab.New("") // KAAB_API_KEY
body, _ := kaab.Body[client.CreateJobJSONBody](map[string]any{
    "url":     "https://shop.example.test/p/1",
    "extract": map[string]any{"title": "h1"},
})
job, _ := c.Jobs.Create(ctx, body, kaab.WithWait(30))
fmt.Println(*job.Data)

curl https://api.kaab.ist/v1/jobs \
  -H "Authorization: Bearer $KAAB_API_KEY" \
  -H "Prefer: wait=30" \
  -d '{"url":"https://shop.example.test/p/1","extract":{"title":"h1"}}'

[Plans]

Start free.Grow by credits.

[How it compares]

Not anotherscraping API.