AI software testing

Your AI writes the code.Testing is now the slow part.

qarunbook lets the same assistant draft the test plan from your code, read what's failing, record results and resolve what it fixes — while a person still confirms every fix before it counts as a pass.

Start testing freeConnect your AI

The MCP connection and both skills are on every plan, including Free.

01Plan
From code

/test-checklist reads your app's code and drafts the runbook, checks and all.

02Connect
14 tools

Over MCP, for Claude Code, Claude Desktop, Cursor, Codex, or any MCP client.

03Access
Per person

Each person connects with their own token and their own role on each app.

04Rule
Fix ≠ pass

A fixed issue waits as a retest until someone checks it again.

Code got faster.Checking it didn't.

AI coding moved the bottleneck. The hard part of a release is no longer writing the change; it's knowing the change works on every platform you ship to.

01Problem

More changes, same testers

An assistant can turn a request into a working branch in an afternoon. The people who check it didn't get faster, so the queue in front of them grew.

02Problem

Nobody wrote the plan

The code arrived before anyone decided what should be tested. Without a list, testing is whatever someone remembers to try.

03Problem

The AI grades its own work

Left alone, an assistant will fix a bug, call it done and report the release clean. Nobody on a real device has seen the fix.

Let the AI draft the plan.From the code it just wrote.

Run /test-checklist in your repo. It reads the app's code, works out every module, feature and user journey worth testing, and creates the app in your runbook with the checks already in it.

Then it's an ordinary plan: edit the wording, add your own checks, choose the platforms, or mark a check not applicable where it doesn't belong.

> /test-checklistExample

Read 214 files. Found 9 modules and 31 journeys. Created Shop app with 3 platforms and 64 checks.

RefCheckWAi
AUTH-01Sign up with email and verify···
AUTH-02Sign in with a magic link···
CART-01Add an item and see the total update···
PAY-01Pay with a saved card···
PAY-02Declined card shows a clear error···
A qarunbook runbook: progress figures, filters, and checks by section, each marked passed, failed or needing a retest on web, Android and iOS, with one check open to show its steps and an issue.

The same runbook your testers work through: checks by section, a result for each platform, and progress you can read at a glance.

Connect your AIto the runbook, over MCP.

14 tools let Claude Code, Claude Desktop, Cursor, Codex or any MCP client list failing checks, record results, and raise, edit and resolve issues.

Everyone connects with their own token, so the assistant can do exactly what that person can on each app, and no more.

Claude Code · connected to qa-runbookExample
>What's failing on Android for Shop app?
⏺list_checks · filter: failing
·Two checks fail on Android. PAY-03: the card form scrolls under the keyboard. AUTH-05: the reset link opens the browser, not the app.
>Fix PAY-03.
·Fixed in CheckoutForm.tsx and committed.
⏺resolve_issue · “Card form now resizes with the keyboard.”
·PAY-03 now reads as retest on Android. Someone needs to check it on a device before it counts as a pass.
Terminal · connect Claude Code
claude mcp add qa-runbook --transport http https://qarunbook.com/api/mcp \
  --header "Authorization: Bearer <your token>"
Terminal · add both skills as a plugin
claude plugin marketplace add Ifeanyiejindu/qarunbook
claude plugin install qarunbook@qarunbook

Your token is under Connect your AI once you have a workspace, with the lines for Claude Desktop, Cursor and Codex beside it. More in the MCP server guide.

01Read · 5
  • list_apps List appsany role
  • progress Coverage summaryany role
  • list_issues List issuesany role
  • list_checks List checksany role
  • get_check Get one checkany role
02Add · 5
  • create_app Create an appadmin
  • import_plan Import a test planadmin
  • add_section Add a sectionadmin
  • add_check Add a checkadmin
  • add_issue Raise an issuetester
03Change · 4
  • resolve_issue Mark an issue fixedadmin
  • edit_issue Edit an issuetester
  • set_result Record a resulttester
  • update_check Correct a checkadmin

Then let it helpwith the testing itself.

/qa-runbook is the second skill. It works through the plan and writes back what it finds, so the results sit beside your testers' in the same runbook.

01/qa-runbook

Establishes what is under test

Works out which app and which platforms it is testing before it records anything.

02/qa-runbook

Drives web and mobile

Works through the checks on the platforms the app ships to, with an account you give it or a test account it creates.

03/qa-runbook

Records each result with evidence

Pass or fail per platform, only for what it actually saw, and an issue on the check when something breaks.

04/qa-runbook

Verifies fixes

Goes back to the checks waiting on a retest and records what it finds on the new build.

A fix is not a passuntil someone checks again.

When an issue is resolved, over MCP or on the web, the check doesn't go back to passing. It reads as retest on the platforms the issue affected, because the last result was recorded on a build without the fix.

The retest clears only when a new result is recorded after the fix. Every result keeps who recorded it and when.

PAY-03 · AndroidExample
  1. F
    A tester raises an issue
    Open issue: the check reads as failed on the platforms it affects.
  2. R
    The AI fixes it and resolves the issue
    Retest. The old pass was recorded on a build without the fix.
  3. P
    A person checks the new build
    A result recorded after the fix clears the retest.
One check opened, showing its preconditions, steps, expected result and an open issue on Android.

Issues stay with their checks

A tester's issue lands on the check that found it, with the platforms it affects and any screenshots or recordings. That's what your assistant reads when you ask it what's broken.

Free to start.Flat pricing for teams.

The MCP connection and both skills are on every plan. Paid plans raise the limits and add the REST API, at one price per workspace — never per seat.

The full breakdown and its FAQ are on the pricing page.

Free
$0/mo

For one person or a small team trying qarunbook on a real release.

  • 2 apps
  • Up to 5 members
  • 2 GB of screenshots and recordings
  • MCP connection and both AI skills
  • Roles, join links and the retest flow
Most popular
Team
$29/mo

For a team testing several apps every release.

  • 15 apps
  • Up to 15 members
  • 50 GB of screenshots and recordings
  • MCP connection and both AI skills
  • Roles, join links and the retest flow
  • API access
Business
$79/mo

For QA across a whole portfolio of apps and clients.

  • Unlimited apps
  • Up to 50 members
  • 250 GB of screenshots and recordings
  • MCP connection and both AI skills
  • Roles, join links and the retest flow
  • API access
Enterprise
Custom

Unlimited everything, invoiced, with terms that suit your procurement.

  • Unlimited apps and members
  • Unlimited storage
  • MCP connection and both AI skills
  • Invoiced billing
  • Talk to us at hello@qarunbook.com
  • API access

Prices in USD, before tax. Tax is added only where we’re registered to collect it. Cancel anytime.

Questions,answered.

The API docs and the Claude Code skill guide go further. Building apps for clients? See qarunbook for agencies.

Does qarunbook run my tests?
No. qarunbook holds the plan, the results and the follow-up. People test, and assistants connected over MCP read the plan and record what they find. The /qa-runbook skill is what tells an assistant how to do the testing.
Which assistants work with it?
Anything that speaks MCP over HTTP. Claude Code, Cursor and Codex each install it as one plugin from the GitHub repo, carrying the connection and both skills; Claude Desktop connects with a custom connector. Any other agent that reads SKILL.md files takes the skills with npx skills add Ifeanyiejindu/qarunbook.
Can the AI mark its own fix as passed?
Resolving an issue never passes a check: it reads as retest until a new result is recorded after the fix. An assistant working with a tester's token can record a result, the same as that person could, which is why every result keeps who recorded it and when, and why the /qa-runbook skill tells the assistant never to record a pass it did not observe.
What can an assistant do with a viewer's token?
Read. A token carries its person's role on each app, so a viewer's assistant can list checks, issues and progress but can't change anything. Recording results and raising issues need tester access; editing the plan and resolving issues need admin.
Is it free?
The MCP connection and both skills are on every plan, including Free. Paid plans raise the limits on apps, members and storage, and add the REST API.
Is there an API as well?
Yes, on paid plans: a REST API with workspace keys, for raising issues from a support desk or reporting results from CI. See qarunbook.com/docs.

Let your AI write the plan.Keep a person on the last check.

Create a workspace, connect your assistant, and run /test-checklist on your app.

Start testing freeRead the docs

Free plan available — no card required.

Works with your AI and every platform you ship to

Claude
Codex
Cursor
Any MCP client
Web
Android
iOS