qarunbook 通过 MCP 让 Cursor 读取测试人员上报的缺陷并修复

Original title: QA runbook over MCP: Cursor reads tester-reported bugs, fixes them, and sends them back for a human retest

The title and summary in the selected language are awaiting translation.

AI overview

作者 Ifeanyi Ejindu 发布 qarunbook，一个面向 Web 和移动应用的人工测试 runbook，用检查项（步骤加预期结果）、各平台通过/失败状态和针对检查项提出的 issue 组织测试，不跑自动化测试。

Full text

2026 年10 月 2 日 14:00 1

Hi all, I’m Ifeanyi, and I build qarunbook. Sharing it here because it’s built to be used from Cursor.

What it is: a manual-testing runbook for web and mobile apps. Test plans of checks (steps + expected result), a pass/fail per platform, and issues raised against the check that found them. It doesn’t run automated tests.

The Cursor part: an MCP server at https://qarunbook.com/api/mcp (Bearer token). Once it’s added in mcp.json, the agent can:

list failing checks and open issues (“What’s failing on Android?”)
fix the ones in your repo and mark each issue fixed with a note
import a Markdown test plan and point out what it doesn’t cover

One design decision I’d like feedback on: marking an issue fixed sends the check to retest. It doesn’t pass it. The agent can’t record a pass for its own fix. Is that the right call, or would you want an override for trivial fixes?

Configs for Cursor, Claude Code and Codex: QA MCP server — qarunbook

Free plan: 2 apps, 5 people, no card.