qarunbook

Work out what needs testing, then actually get it tested.

Most teams know they should test before a release. What stops them is not discipline — it is that nobody wants to sit down and write out three hundred things to check, and once somebody has, there is nowhere sensible to record what happened to each one.

qarunbook does both halves:

1. Itemise. Point an AI assistant at your codebase and it works out every module, feature and user journey worth testing, across each platform you ship on, and writes them as checks a non-technical person can follow. Or write them yourself, or upload a plan you already have.
2. Test and track. Testers work top to bottom, marking each check passed or failed and raising what breaks. You can see how far along the release is at any moment, and what is in the way.

It is built for user acceptance testing — the pass a human makes over a real build before it ships. Not unit tests, not CI. The kind of testing where somebody opens the app, signs up, and tries to actually use it.

https://qarunbook.com — free to use.

Why an MCP server

Because the loop usually breaks at the hand-off. A tester finds a bug, writes it in a spreadsheet or a WhatsApp message, and it is now somebody's job to read that, understand it, fix it, and remember to tell the tester it is done.

qarunbook exposes the runbook over MCP, so your coding assistant is on the other end of that loop directly:

list_issues            what did the testers actually find (and what they attached)
list_checks failing    what is broken right now
edit_issue             that report was vague — here is what actually happens
resolve_issue          I fixed it

Testers can attach screenshots, screen recordings and logs to an issue in the web app. Videos stream straight in the page, so a bug is something you watch rather than something you read about.

When an issue is marked fixed, the check does not silently go back to passing. It reads R — retest, because the person who last passed it was looking at a build without the fix in it. It clears when somebody actually looks again. Status is derived from marks and issues rather than stored, so the runbook can never quietly disagree with itself.

Install

Pick your assistant. For Claude Code, Cursor and Codex the repo is a plugin marketplace, so one install brings the MCP connection and both skills; the only thing you add by hand is your token, from your qarunbook home page under Connect your AI. The token is yours, not the workspace's. It acts as you, with your role, and removing you revokes exactly it.

Claude Code

/plugin marketplace add Ifeanyiejindu/qarunbook
/plugin install qarunbook@qarunbook

(or from a terminal: claude plugin marketplace add Ifeanyiejindu/qarunbook then claude plugin install qarunbook@qarunbook)

Then set your token:

export QARUNBOOK_TOKEN="qarb_…"

Cursor

Cursor installs plugins, not loose skill folders, so the repo carries a .cursor-plugin marketplace manifest. In Cursor:

1. Open Customize in the sidebar and choose From GitHub Repository (or Plugins → Add Marketplace).
2. Paste https://github.com/Ifeanyiejindu/qarunbook.
3. Install qarunbook.
4. Under Plugins → Configure, set QARUNBOOK_TOKEN to your token.

That is the whole install: the qa-runbook MCP server and both skills arrive together.

Codex

The repo is also a Codex plugin marketplace (.agents/plugins/marketplace.json). In the shell that launches Codex:

export QARUNBOOK_TOKEN="qarb_…"
codex plugin marketplace add Ifeanyiejindu/qarunbook

Then inside Codex run /plugins, find qarunbook and install it. Codex reads the token from the environment when it connects and sends it as a bearer token. Update later with codex plugin marketplace upgrade.

Other agents — Gemin

... View full README on GitHub