Skip to content

pencil20388-eng/browser-automation-skills / skills/chrome-devtools-mcp-adspower

chrome-devtools-mcp-adspower

Verified

Chrome DevTools MCP + AdsPower Integration Skill

Connect Google's Chrome DevTools MCP to AdsPower antidetect browser profiles. This lets AI coding agents (Claude Code, Cursor, Codex CLI) control fingerprint-isolated browser sessions through natural language.

Why This Combination

  • Chrome DevTools MCP gives AI agents eyes in the browser: inspect DOM, monitor network, run JavaScript, take screenshots
  • AdsPower gives each browser session a unique fingerprint, proxy, and identity
  • Together: your AI agent can operate across multiple isolated browser identities — each with its own fingerprint, cookies, and IP address
Claude Code → Chrome DevTools MCP → AdsPower Profile (unique fingerprint + proxy)
                                  → AdsPower Profile (unique fingerprint + proxy)
                                  → AdsPower Profile (unique fingerprint + proxy)

How It Works

AdsPower's Local API opens browser profiles and exposes a Chrome DevTools Protocol (CDP) endpoint for each one. Chrome DevTools MCP connects to that endpoint via --browserUrl.

Prerequisites

  • AdsPower installed and running (paid plan with API access)
  • Chrome DevTools MCP installed (npm install -g chrome-devtools-mcp@latest)
  • API Key generated (AdsPower → Automation → API → Generate)
  • Python 3.8+ with requests installed

Step 1: Open an AdsPower Profile and Get the CDP Endpoint

import requests

API_KEY = "your_api_key"
BASE_URL = "http://local.adspower.net:50325"
HEADERS = {"Authorization": f"Bearer {API_KEY}"}

PROFILE_ID = "your_profile_id"

# Open the browser profile
resp = requests.get(
    f"{BASE_URL}/api/v1/browser/start?user_id={PROFILE_ID}",
    headers=HEADERS,
    timeout=30,
).json()

if resp["code"] == 0:
    # This is the CDP WebSocket endpoint
    ws_endpoint = resp["data"]["ws"]["puppeteer"]
    debug_port = ws_endpoint.split(":")[2].split("/")[0]
    print(f"CDP WebSocket: {ws_endpoint}")
    print(f"Debug port: {debug_port}")
else:
    print(f"Error: {resp['msg']}")

The ws_endpoint looks like: ws://127.0.0.1:XXXXX/devtools/browser/GUID

Step 2: Connect Chrome DevTools MCP to the AdsPower Profile

Once the profile is open and you have the debug port, connect Chrome DevTools MCP:

# Connect to the AdsPower browser instance
npx chrome-devtools-mcp@latest --browserUrl=http://127.0.0.1:XXXXX

Replace XXXXX with the actual port from Step 1.

Configure in Claude Code

Add to your MCP config (.claude/mcp.json or settings.json):

{
  "mcpServers": {
    "chrome-devtools": {
      "command": "npx",
      "args": [
        "-y",
        "chrome-devtools-mcp@latest",
        "--browserUrl=http://127.0.0.1:XXXXX"
      ]
    }
  }
}

Step 3: Use Natural Language to Control the Browser

Once connected, you can tell your AI agent:

> Navigate to amazon.com and check if the account is still logged in

> Take a screenshot of the current page

> Check the network requests for any blocked or failed calls

> Run document.title in the browser console and tell me what page this is

> Inspect the fingerprint by visiting browserscan.net

The AI agent sees the browser through Chrome DevTools MCP, while AdsPower ensures the browser has a unique fingerprint and proxy.

Automating Multiple Profiles

To work across multiple AdsPower profiles sequentially:

import requests
import subprocess
import time

API_KEY = "your_api_key"
BASE_URL = "http://local.adspower.net:50325"
HEADERS = {"Authorization": f"Bearer {API_KEY}"}

def get_profiles(group_name):
    resp = requests.get(
        f"{BASE_URL}/api/v1/user/list",
        headers=HEADERS,
        params={"group_name": group_name, "page": 1, "page_size": 100},
    ).json()
    return resp["data"]["list"]

def open_profile(profile_id):
    resp = requests.get(
        f"{BASE_URL}/api/v1/browser/start?user_id={profile_id}",
        headers=HEADERS,
        timeout=30,
    ).json()
    if resp["code"] == 0:
        ws = resp["data"]["ws"]["puppeteer"]
        port = ws.split(":")[2].split("/")[0]
        return port
    return None

def close_profile(profile_id):
    requests.get(
        f"{BASE_URL}/api/v1/browser/stop?user_id={profile_id}",
        headers=HEADERS,
    )

# Process each profile
profiles = get_profiles("My Campaign")
for p in profiles:
    port = open_profile(p["user_id"])
    if port:
        print(f"Profile {p['name']} ready on port {port}")
        print(f"Connect MCP: npx chrome-devtools-mcp@latest --browserUrl=http://127.0.0.1:{port}")
        # Your automation here...
        input("Press Enter when done with this profile...")
        close_profile(p["user_id"])
        time.sleep(2)

Use Cases

ScenarioWhat happens
Multi-account health checkOpen each profile → MCP checks login status, screenshots dashboard
Ad verificationOpen profiles with different geo proxies → MCP inspects ad content per region
Fingerprint auditingOpen profiles → MCP navigates to BrowserScan → reads fingerprint results
Authenticated scrapingOpen profile with saved cookies → MCP extracts data from logged-in pages
A/B test across identitiesOpen multiple profiles → MCP captures different UI versions

Important Notes

  • Each AdsPower profile exposes its own CDP port — you can only connect one MCP instance per profile at a time
  • Close the MCP connection before closing the AdsPower profile to avoid orphaned processes
  • The CDP port changes every time a profile is opened — always read it fresh from the API response
  • AdsPower handles fingerprint and proxy; Chrome DevTools MCP handles browser control and inspection
  • Rate limit: wait 1-2 seconds between opening/closing profiles

Resources

reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall