Browser Testing With Devtools

Browser Testing with DevTools

I'll connect to a running browser session and inspect it at runtime: DOM state, console errors, network traffic, and rendering performance. Requires Chrome DevTools MCP to be configured.

Arguments: $ARGUMENTS — URL to open or test to perform (e.g., "check login flow at http://localhost:3000")

Token Optimization

Expected range: 300–800 tokens per inspection (snapshot + analysis)

Early exit: If no issues are found in the initial snapshot, report "No issues detected" and stop

Patterns used: Progressive disclosure (summary first, full DOM/network only on specific request)

Prerequisites

Chrome DevTools MCP must be configured. Verify:

# Check if chrome-devtools-mcp is in MCP config
grep -r "devtools\|chrome" ~/.claude/settings.json .claude/settings.json 2>/dev/null | head -5

If not configured, report the requirement and stop.

Step 1 — Open or Connect to Page

Navigate to the target URL:

Navigate to: [URL from $ARGUMENTS or inferred from project]
Wait for: DOM ready / network idle

Take an initial snapshot of the page state (accessibility tree + title).

Step 2 — Console Error Scan

Capture all console messages:

Console output:
  errors:   [count]
  warnings: [count]
  logs:     [count]

Top errors (if any):
  [message] — [source file:line]

Any console.error or unhandled promise rejection is a test failure. Report immediately.

Step 3 — DOM Inspection

For the specific flow being tested, inspect the relevant DOM elements:

• Verify expected elements exist and are visible
• Check ARIA roles, labels, and focus order for accessibility
• Confirm form inputs, buttons, and interactive elements are in the expected state
• Flag any elements with display: none, visibility: hidden, or opacity: 0 that should be visible

DOM check:
  [element] — [found / not found / wrong state]
  [element] — [found / not found / wrong state]

Step 4 — Network Request Analysis

Review network requests made during the test flow:

Network:
  total requests: [N]
  failed (4xx/5xx): [N]
  slow (>500ms): [N]

Failed requests:
  [URL] — [status] — [response body excerpt]

Flag any unexpected 404s, 401s, CORS errors, or requests to the wrong environment.

Step 5 — Performance Snapshot

For pages where rendering performance matters:

Performance:
  FCP (First Contentful Paint): [N]ms — [good 3s]
  LCP (Largest Contentful Paint): [N]ms — [good 4s]
  Layout shifts: [N] — [stable / unstable]

Flag any Core Web Vitals that are in the "needs work" or "poor" range.

Step 6 — Report

Browser Test: [URL] — [PASS / FAIL]

Issues found: [N]
  [severity] [category]: [description]
  ...

Passed checks:
  - [check description]
  ...

Edge Cases

• Chrome DevTools MCP not available: report the requirement; suggest using Playwright for automated testing instead
• Page requires authentication: note the requirement; the user must provide a session cookie or perform login steps manually before calling this skill
• Single-page app not fully hydrated: wait for network idle before inspecting; add explicit wait if hydration takes more than 2 seconds
• Local dev server not running: detect via connection refused; report which server to start

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: manastalukdar
• Source: manastalukdar/ai-devstudio
• License: MIT

Install and usage instructions live in the source repository linked above.