Web App QA Loop
For checking a web app the way a person would, without clicking through it yourself. Playwright's MCP server lets the agent drive a real browser: open pages, click, fill forms and read the page. Chrome DevTools' MCP server shows what the page doesn't: console errors, failed network requests and performance traces. When something breaks, the Test-Driven Development skill has the agent write a test that fails on the bug before it changes any code, and Anthropic's Web App Testing skill turns the checks into Playwright scripts you can run again. squirrelscan's Audit Website skill then sweeps the whole site for SEO, performance, security and accessibility problems, fixes them in batches and re-audits until the site reaches its target score. GitHub's MCP server files what's left as issues, or opens the pull request.
Skills
Web App Testing
Writes Python Playwright scripts to test a local web app, with a helper that starts and stops your dev servers around each run. Screenshots, DOM inspection and browser logs.
npx skills add anthropics/skills@webapp-testingTest-Driven Development
No fix without a failing test first: the agent writes the test, watches it fail on the bug, then writes the smallest change that makes it pass.
npx skills add obra/superpowers@test-driven-developmentMCP Servers
Playwright
Microsoft's browser automation server. The agent reads pages as structured accessibility snapshots rather than screenshots, and clicks, types and navigates through them.
Chrome DevTools
Google's server for inspecting a live Chrome: console messages, network requests, performance traces and screenshots. It can read and change anything in that browser, so don't sign it into accounts you wouldn't share with your agent.
GitHub
Files the issues the run found, or opens the pull request with the fix.
Setup
Three skills, in the project you're testing. Web App Testing writes Python, so it needs Python with the Playwright package and a browser installed.
npx skills add anthropics/skills@webapp-testing
npx skills add obra/superpowers@test-driven-development
npx skills add squirrelscan/skills@audit-websitePlaywright drives the browser; Chrome DevTools inspects it. Chrome DevTools sends usage statistics to Google by default. Its README documents the --no-usage-statistics flag and the CHROME_DEVTOOLS_MCP_NO_USAGE_STATISTICS variable to turn that off, and --no-performance-crux to stop performance tools sending page URLs to Google's CrUX API.
claude mcp add playwright npx @playwright/mcp@latest
claude mcp add chrome-devtools --scope user npx chrome-devtools-mcp@latestThe local server runs in Docker with a personal access token. Give the token only the repository you're testing and the issue and pull-request scopes the agent needs.
claude mcp add github -e GITHUB_PERSONAL_ACCESS_TOKEN=$GITHUB_PAT -- docker run -i --rm -e GITHUB_PERSONAL_ACCESS_TOKEN ghcr.io/github/github-mcp-serverAudit Website runs the squirrel command. Signed out, an audit runs on your machine for free and nothing is published; signed in, every audit is billed in credits, and the skill tells the agent to show you the cost and get a yes first. Its frontmatter pre-approves squirrel commands and file edits, so in the turn you invoke it, those run without asking.
npm install -g squirrelscanStart your dev server and ask the agent to walk the main flows in the browser and report anything broken, with the console and network errors behind it. For each bug, have it write a failing test first, then fix it. Then run an audit, approve the fixes you want, and have it re-audit. Ask it to file what you didn't fix as GitHub issues.
Use Cases
- Checking the main flows of a web app before a release
- Turning a bug you found by hand into a failing test and a fix
- Finding the console and network errors behind a broken page
- Cleaning up SEO, accessibility and performance problems across a site, with a before and after score
Get new workflows in your inbox. Skills and MCP servers combined for a whole job, one a week. See a past issue.