Overview

Nitpick is an AI QA team. A Senior QA agent coordinates. Junior QA agents do the deep work per page or flow. You get a single unified report, a persistent knowledge base, and — in v2.0+ — a live MCP server your IDE can query at any time.

What Nitpick does

• Discovers your app: crawls as each role, handles SPAs (deterministic + LLM-assist), builds a page graph
• Decides scope: you describe a change; Nitpick proposes which pages are affected and asks you to confirm
• Tests each page deeply: per-page subagent explores the UI, builds a Derived UI Model, generates Playwright tests, runs them, classifies failures, retries intelligently
• Tests multi-actor flows: admin creates → candidate approves → admin finalizes, with cross-actor assertions
• Smoke-checks the rest: quick regression check on critical pages not in scope
• Reports: unified HTML dashboard, markdown summary, JSON for dashboards
• Remembers: persistent knowledge base of page models, flaky tests, known bugs that survives across runs
• Exposes the KB via MCP: nitpick mcp serve makes the knowledge base queryable from Claude Code, Cursor, Cline, and any MCP client

What Nitpick is not

• Not a hosted SaaS. You run it. Your data stays with you.
• Not a test generator that writes tests once and forgets. Nitpick is a continuous QA team.
• Not Claude Code. Nitpick has its own agent runtime, supports Anthropic, OpenAI, and optionally Claude Code. The MCP server extends into Claude Code without depending on it.

Start here

• Install: requirements, 4-command install, Playwright setup
• Quickstart: your first end-to-end run
• Your first run: a walkthrough of what you’ll see the first time
• MCP KB server: connect Claude Code, Cursor, or Cline to your KB
• QA team skills: the 20 skills that cover every QA role
• Architecture: how all the pieces fit, including the MCP dual-surface design

Get help

• GitHub: vaibhav-kindflow/nitpick
• Issues and questions: file on GitHub