Skip to content

/inspect-in-devtools

Use this skill to connect to the Chrome DevTools UI for interactive debugging, UI/UX review, Q&A, showing elements/network requests/panels to the user, inspecting the user's current DevTools selection, and addressing or polling for DevTools UI comment threads.

From plugin
chrome-devtools-mcp
56k8 skills

Claude Code

$ npx -y skills add ChromeDevTools/chrome-devtools-mcp --skill inspect-in-devtools --agent claude-code

SKILL.md

inspect-in-devtools.SKILL.md
name: inspect-in-devtools
description: Use this skill to connect to the Chrome DevTools UI for interactive debugging, UI/UX review, Q&A, showing elements/network requests/panels to the user, inspecting the user's current DevTools selection, and addressing or polling for DevTools UI comment threads.

Debug, Review, and Q&A in DevTools

Use this skill to:

1. **Show a target to the user**: Open DevTools and reveal a panel, a DOM element, or a network request. 2. **Inspect the user's selection**: Read the element or network request that the user selected in DevTools. 3. **Address DevTools comments**: Fetch comment threads, answer questions or apply fixes, and resolve the threads.

Prerequisites

  • **Chrome version**: Chrome M159 or later.
  • **Required flag**: `--devtools-comments` in the MCP server config. Without it, `open_devtools`, `reveal_in_devtools`, `get_devtools_comments`, and `resolve_devtools_comment` are not available.
  • **Headed browser**: Chrome must not run in `--headless` mode.
  • **Page targeting**: All tools require a `pageId` from `list_pages` or `new_page`.

Workflows

1. Open DevTools

Call `open_devtools` with `pageId`. Do this before you call `reveal_in_devtools`, `get_devtools_comments`, or `resolve_devtools_comment`.

2. Reveal a panel, element, or network request

Call `reveal_in_devtools` with `pageId` and these optional parameters:

  • `panelName`: `"elements"`, `"network"`, `"sources"`, or `"console"`. Pass it alone to switch panels.
  • `uid`: Element `uid` from `take_snapshot`. Reveals the element in the Elements panel.
  • `reqid`: Request `reqid` from `list_network_requests`. Reveals the request in the Network panel.

`uid` and `reqid` are mutually exclusive.

3. Inspect the user's selection

  • **DOM element**: Call `take_snapshot`. The snapshot marks the element selected in the Elements panel. Pass its `uid` to `get_css_styles`, `take_screenshot`, or `evaluate_script`.
  • **Network request**: Call `get_network_request` and omit `reqid`. It returns the request selected in the Network panel.

4. Handle comment threads

Select the mode from the user's request.

Mode A: One-shot (comments already exist)

Use this mode when the user says the comments are already in DevTools, or asks you not to poll.

1. Call `get_devtools_comments` once. 2. Process and resolve each returned thread. 3. Summarize your changes and answers in chat, then end your turn. Do **not** poll.

Mode B: Watch (wait for user feedback)

Use this mode when the user asks you to wait while they debug, review, or add comments.

1. Call `get_devtools_comments` immediately. 2. While the result is `"No open DevTools comments found."`, wait 10 seconds, then call it again. Call it immediately if a `"DevTools comment threads updated"` notification arrives. 3. Stop after 12 consecutive empty polls (about 2 minutes) or when the user asks you to stop. 4. When threads appear, process and resolve them. Then reset the empty-poll counter and continue polling.

Rules for polling:

  • **Use a background sub-agent** when sub-agent tools are available, so the main thread stays responsive. The sub-agent returns the threads to the main agent, or resolves them itself if it has workspace and MCP tools.
  • **Poll inline** only when sub-agent tools are unavailable. Do not end your turn early.
  • **Never call `get_devtools_comments` back-to-back** without the 10-second delay.

Process and resolve a thread

For each `### Thread: <id>`:

1. **Inspect the anchored target**:

  • `Related element uid`: Use `get_css_styles`, `take_snapshot`, or `take_screenshot` with the `uid`.
  • `Related network reqid`: Use `get_network_request` with the `reqid`.
  • Optionally call `reveal_in_devtools` to show the target to the user.

2. **Address the comment**:

  • For a code or styling request, apply the fix in the workspace. Reload and verify the page if needed.
  • For a question or root-cause investigation, collect diagnostic data. Do not make unnecessary code edits.

3. **Resolve**: Call `resolve_devtools_comment` with `pageId`, `threadId`, and `replyText`:

  • For a question or root cause, write a detailed answer.
  • For a straightforward fix, write 1–2 sentences that state the change and the file.
Read more
@chromedevtools
Ships withchrome-devtools-mcp

Chrome DevTools for agents (chrome-devtools-mcp) lets your coding agent (such as Antigravity, Claude, Cursor or Copilot) control and inspect a live Chrome browser. It is available as a Model-Context-Protocol (MCP) server and as a CLI.

Stars on GitHub

Momentum

stars in total
56.4k
Star history of ChromeDevTools/chrome-devtools-mcp on GitHub, a chart by star-history.com

Other skills on chrome-devtools-mcp.

More skills from other plugins.

cpu-high

cpu-high

排查 JVM / 应用 CPU 飙高(线程定位 + 代码路径分析)

arthas6d38k
@alibaba@alibabaInstall Skill
analyze-dump

analyze-dump

Triage a Windows crash dump - identify the exception, the faulting frame, and what to look at next. Use when the user points at a…

@svnscha@svnschaInstall Skill
d3k

d3k

Test and debug a web app in its existing d3k browser, including local Chrome connected to a Vercel Sandbox over CDP. Use for UI…

dev30005d1.6k
@vercel-labs-2@vercel-labs-2Install Skill
gearboy-debugging

gearboy-debugging

Debug and trace Game Boy / Game Boy Color / Super Game Boy games using the Gearboy emulator MCP server. Provides workflows for…

gearboy2d1.2k
@drhelius@drheliusInstall Skill

Questions about inspect-in-devtools.

Which coding agents can use inspect-in-devtools?
inspect-in-devtools installs in Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, OpenCode and Windsurf with the skills CLI, using each agent's flag shown on this page. Flowy's auto-invocation, which fires a skill when your prompt matches, runs in Claude Code only.