Skip to content

Summary

Give coding agents your real Chrome session: console logs, network traffic, screenshots and Lighthouse audits streamed through a DevTools extension, with credentials scrubbed and a loopback-only connector.

Features

  • Attaches to your real, logged-in Chrome profile via a DevTools extension
  • Console and network logs with keyword filters, paging and HAR export
  • Screenshots returned as images and selected-element inspection
  • Lighthouse accessibility, performance, SEO and best-practices audits
  • Per-tab tracking with explicit tabId targeting
  • Loopback-only connector, per-run token and credential scrubbing (v2)

Installation

Set up this MCP server in your favorite AI agent — copy a ready-made configuration below.

Any MCP-compatible agent

Most agents (Claude, Cursor, Windsurf, VS Code, and more) read a standard mcpServers configuration.

  1. Open your agent's MCP configuration file.
  2. Merge the snippet below into it.
  3. Restart the agent — the "BrowserTools MCP" tools will be available.
{
  "mcpServers": {
    "browser-tools-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@agentdeskai/browser-tools-mcp@latest"
      ]
    }
  }
}

Description

BrowserTools MCP from AgentDesk lets a coding agent see what is happening in the browser you are already using. A Chrome DevTools extension captures console output, network requests, screenshots and the selected element from your normal, logged-in profile and passes them to the MCP server, so Cursor, Claude Code, Windsurf, Cline, Zed or Gemini CLI can answer "why is checkout throwing?" against the actual page.

The README is explicit about where it fits: CDP-driven servers like Playwright MCP or Chrome DevTools MCP launch a fresh automated browser, which is right for tests, while BrowserTools attaches to your existing session — useful because Chrome refuses remote debugging on the default profile that holds your logins.

Tools

Console logs and errors with keyword filters, network logs and failures (with timing and bodies), selected element, page info, tab listing and per-tab targeting, screenshots returned as images, page refresh, storage and cookies (values gated), log wiping, light navigation and soft click/type, plus Lighthouse accessibility, performance (Core Web Vitals), SEO and best-practices audits. Large payloads such as full logs, HAR files and complete audit reports are exposed as MCP resources rather than inlined, and every read tool pages with limit/offset.

Version 2 and security

2.0 is a rewrite: one process instead of three, a connector bound to 127.0.0.1 only, a per-run bearer token, Authorization/Cookie headers and tokens redacted before they leave the browser, and SSRF checks on audits. The README says 1.x had critical vulnerabilities and should be upgraded.

Setup

Add npx -y @agentdeskai/browser-tools-mcp@latest to your MCP client (Node 22.19+), then load the chrome-extension folder from the repository unpacked at chrome://extensions. Open DevTools on the page to start capture; --doctor diagnoses a broken install. MIT licensed.

MCP: Browser MCP

by Browser MCP

New

Lets Claude Code, Cursor, VS Code, Codex and other AI apps drive your own Chrome, with your signed-in sites and profile, through a local MCP server and a companion extension.

MCP: Canvas MCP

by Vishal Sachdev

Connects AI assistants to Canvas LMS with up to 101 tools covering assignments, grading, analytics, discussions, accessibility auditing and course structure.

Open-source MCP server that brings Google Search Console into an AI assistant — query clicks, impressions, CTR and position, compare periods, inspect URL indexing and manage sitemaps.

Browse all MCP servers →