Skip to content

Chrome DevTools MCP Skills - Setup Guide

Source: chromedevtools/chrome-devtools-mcp GitHub: chromedevtools/chrome-devtools-mcp Skills: chrome-devtools (5,934 installs) · chrome-devtools-cli (2,572 installs) Category: Browser Automation & Debugging First Seen: August 14, 2026 sweep Quality Tier: 🟢 Production

The official Google Chrome DevTools skills let an agent drive a real Chrome browser through the Chrome DevTools Protocol (CDP) - navigate pages, click and fill forms via accessibility-tree snapshots, read console output and network requests, capture performance traces, and debug JavaScript. Because the transport is an MCP server (npx chrome-devtools-mcp), any MCP-capable agent can use it, including Hermes via its native MCP client. Two skills ship in the repo: chrome-devtools for agent-driven tool calling, and chrome-devtools-cli for browser automation from shell scripts.

Installation

Install both skills from skills.sh:

npx skills add chromedevtools/chrome-devtools-mcp

Register the MCP server in Hermes (~/.hermes/config.yaml under mcp_servers):

mcp_servers:
  chrome-devtools:
    command: npx
    args: [chrome-devtools-mcp@latest]

Optional capability flags (add to args): - --categoryExtensions - enable extension tooling (install/list/inspect extensions) - --memoryDebugging - enable memory debugging tools - --browserUrl http://127.0.0.1:9222 - attach to an existing Chrome instead of auto-launching

The server starts Chrome automatically on first tool call using a persistent profile. The skill does not apply in --slim mode (no MCP configuration).

Prerequisites

Requirement Details
Node.js 20.19+ / 22.12+ Required by chrome-devtools-mcp (Node version check enforced)
Google Chrome Stable or Canary - the server launches it automatically on first call
Hermes Agent Any recent version with the native MCP client (see hermes mcp / config.yaml mcp_servers)
npx Bundled with Node.js - used to run the MCP server

What It Provides

chrome-devtools (MCP tools, agent-driven)

Capability Tools Notes
Page navigation navigate_page, new_page, list_pages, select_page Tools operate on the currently selected page
Interaction take_snapshot, click, fill, hover, press_key Snapshot returns uid refs - always re-snapshot before acting
Script evaluation evaluate_script Run JS in the page, return JSON
Network inspection list_network_requests, get_network_request Request/response headers and bodies
Console & errors list_console_messages, get_console_message Catch client-side errors
Performance performance_start_trace, performance_stop_trace Core Web Vitals + insight highlights
Extension tooling install_extension, list_extensions (requires flag) For testing your own extensions

chrome-devtools-cli (terminal automation)

chrome-devtools list_pages
chrome-devtools take_snapshot          # uid-based element snapshot
chrome-devtools click --uid 1_3
chrome-devtools fill --uid 1_4 --value "hello@example.com"
chrome-devtools navigate_page --url https://www.corpusiq.io/docs/
chrome-devtools evaluate_script --function "() => document.title"

The CLI starts the background server implicitly - do not run start/status/stop before each use. State persists across commands.

Quick Start

  1. npx skills add chromedevtools/chrome-devtools-mcp
  2. Add the mcp_servers.chrome-devtools block to ~/.hermes/config.yaml and restart Hermes
  3. In session: "list my open Chrome pages and take a snapshot of the active one"
  4. "Run an accessibility snapshot of https://www.corpusiq.io/docs/ and click the pricing link"
  5. For scripted work: chrome-devtools take_snapshot | grep "Sign in"

CorpusIQ Use Cases

Use Case How
Docs site QA Take snapshots and screenshots of docs.corpusiq.io pages after deploys to verify rendering, catch console errors
Frontend debugging Inspect console messages and network requests on www.corpusiq.io to trace failed API calls
Form-flow testing Automate signup/demo-request flows with click/fill to verify conversion paths end-to-end
Performance auditing Run performance traces against landing pages to catch LCP/CLS regressions before launch
Competitor page teardown Snapshot competitor pricing/dashboard flows headlessly for structured research notes

Integration with CorpusIQ

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

Limitations / Verification

  • File access is restricted to the OS temp dir by default; unrestricted paths need --allowUnrestrictedPaths=true
  • Single browser instance per server - parallel page work must be serialized through the same server
  • MCP tools require a full Hermes restart after config.yaml changes
# Verify the MCP server is registered
hermes mcp list | grep chrome-devtools

# Functional test via CLI
chrome-devtools navigate_page --url "https://example.com" && chrome-devtools take_snapshot

Security

← Skills Catalog | Marketplace →

Powered by CorpusIQ

AI answers on your data. See where every number comes from.

Connect your tools and ask in ChatGPT, Claude, or Perplexity. Enterprise-grade answers in minutes. From $29.95/month.

Start free, no credit card
Contact Terms Privacy Pricing