Skip to content
Aback Inspect

Browser QA for features built with AI

Find the bugs before your users do.

Your coding agent ships features in minutes. Aback Inspect shows whether they work: real Chromium runs on Cloudflare, step by step evidence when something breaks, and an MCP server so the agent can test its own changes before it says it is done.

Start freeRead the quickstart

Free plan with 300 browser runs a month. No card required.

How it works

The loop you already run by hand after every AI change, made repeatable.

  1. Step 1: Add your site

    Production, staging or preview URLs, with the hosts the browser may reach.

  2. Step 2: Describe the flow

    In the visual builder, as JSON, or by asking your coding agent over MCP.

  3. Step 3: Run a real browser

    Chromium on Cloudflare executes every step and records console, network and screenshots.

  4. Step 4: Fix with evidence

    See the failing step, expected and actual values, and the page as it was. Fix and rerun.

Let the agent check its own work

Connect Claude Code, Cursor, VS Code or Windsurf to the Aback Inspect MCP server. After building a feature, the agent writes a test for it, runs it against your preview URL, reads which step failed and why, looks at the failure screenshot, and fixes the code. Then it runs your whole suite so it does not break something else.

  • Tools to list projects, write, validate and run tests, and read results with screenshots.
  • A test guide written for agents, so their first test is a valid one.
  • The same API keys, scopes and quotas as the REST API.
Terminalbash
claude mcp add --transport http aback-inspect \
  https://abackinspect.com/mcp \
  --header "Authorization: Bearer aik_..."
Prompttext
Test the password reset flow you just built on the staging
environment with Aback Inspect. If a step fails, read the run,
fix the code, and run it again until it passes.

Everything a release check needs

One product for functional, visual, crawl and accessibility checks, run on demand, on a schedule, from CI or by an agent.

  • Tests are data, not code

    35 step types: open, click, fill, wait, and assertions on text, URLs, requests, counts and more. Validated on save, versioned on every change, safe to let an agent write.

  • Visual regression

    Pixel comparison against a baseline you accept explicitly, with masks for clocks and ads, side by side and diff views, and history of every baseline.

  • Crawl for broken pages

    Visit linked pages and report broken links, server errors, uncaught exceptions, console errors and missing titles without writing a single step.

  • Console and network checks

    Every run records console errors, uncaught exceptions and failed requests per step, and can fail on them. Query strings are stripped before storage.

  • Accessibility basics

    Automated checks for missing alt text, unlabeled fields, nameless buttons and links, page language and heading order. Basic checks, not a WCAG audit.

  • Schedules and alerts

    Run suites on a timer against verified domains, as often as every 5 minutes on Team. Slack, Discord or signed webhooks when a test starts failing or recovers, not on every run.

  • CI and REST API

    Trigger a suite against a preview URL, wait for it in one request, and get JUnit XML or a Markdown summary for the pull request.

  • Secure by design

    Browsers can only reach your site's allowed hosts, secrets are encrypted with AES-256-GCM and masked in results, and every workspace is isolated.

Readable tests that survive refactors

Steps find elements the way people do: by role and name, label, text or test id. When AI tools rewrite your markup and class names, these tests keep working. When something really breaks, the result says which step, what was expected, and what the page showed instead.

Saving a changed definition creates a new version, every run records the version it used, and assertions wait for the page to settle without ever retrying a failure into a pass.

Honest limits

  • Chromium only. No Firefox, WebKit or real devices.
  • Not a load testing, security scanning or uptime SLA product.
  • Never bypasses CAPTCHAs, paywalls or bot protection.
  • Accessibility checks are basic automated rules, not a WCAG audit.
sign-in.jsonjson
{
  "schemaVersion": 1,
  "kind": "flow",
  "steps": [
    {
      "type": "navigate",
      "path": "/login"
    },
    {
      "type": "fill",
      "target": {
        "by": "label",
        "text": "Email"
      },
      "value": "{{vars.TEST_EMAIL}}"
    },
    {
      "type": "fill",
      "target": {
        "by": "label",
        "text": "Password"
      },
      "value": "{{secrets.TEST_PASSWORD}}"
    },
    {
      "type": "click",
      "target": {
        "by": "role",
        "role": "button",
        "name": "Sign in"
      }
    },
    {
      "type": "expectUrl",
      "match": "/dashboard"
    },
    {
      "type": "expectVisible",
      "target": {
        "by": "role",
        "role": "heading",
        "name": "Your projects"
      }
    },
    {
      "type": "expectNoConsoleErrors"
    }
  ]
}

Simple pricing

Start free. Upgrade for the MCP server, more runs, schedules and alerts.

Free

Verify a side project before every release.

$0/ month

Start free
  • 300 browser runs a month
  • 2 projects, 25 tests
  • 1 run at a time
  • 1 schedule, every day at most
  • 7 days of results and screenshots
  • Crawls of up to 10 pages
  • Visual regression with baselines
  • MCP server for coding agents, REST API and CI
  • Encrypted secrets for test accounts
  • Playwright traces for debugging
  • Slack, Discord and webhook alerts
  • 1 member

Pro

For founders and developers shipping with AI every day.

$19/ month

Start with Pro
  • 3,000 browser runs a month
  • 10 projects, 250 tests
  • 3 runs at a time
  • 20 schedules, every 15 minutes at most
  • 30 days of results and screenshots
  • Crawls of up to 50 pages
  • Visual regression with baselines
  • MCP server for coding agents, REST API and CI
  • Encrypted secrets for test accounts
  • Playwright traces for debugging
  • Slack, Discord and webhook alerts (5 channels)
  • 3 team members

Team

For small teams and agencies with several sites.

$49/ month

Start with Team
  • 15,000 browser runs a month
  • 50 projects, 1,000 tests
  • 10 runs at a time
  • 100 schedules, every 5 minutes at most
  • 90 days of results and screenshots
  • Crawls of up to 100 pages
  • Visual regression with baselines
  • MCP server for coding agents, REST API and CI
  • Encrypted secrets for test accounts
  • Playwright traces for debugging
  • Slack, Discord and webhook alerts (20 channels)
  • 15 team members
Compare plans in detail

Questions

Founders, solo developers and small teams who ship features with Claude Code, Cursor, Copilot or other coding agents, and want proof that each feature works in a real browser before it reaches users. No QA team or browser farm required.

Ship the next feature with proof it works

Sign up, run the sample suite in one click, then point Aback Inspect at your own site.

Start freeBrowse the docs

Questions or a feature request? Contact us