How is Vibe Browser MCP different from Playwright MCP, Chrome DevTools MCP, and BrowserMCP?

Playwright MCP and Chrome DevTools MCP launch a separate browser by default. Playwright MCP can connect to your existing browser either through its Chrome extension mode or through an explicit --cdp-endpoint when Chrome remote debugging is already enabled, but that is still manual setup and it still lacks multi-agent support, Google Workspace tools, credential vault, and sub-agent orchestration. Chrome DevTools MCP requires --remote-debugging-port and sends telemetry to Google. BrowserMCP is a Chrome extension like Vibe but only supports a single agent, has ~13 tools, and its extension is closed-source. Vibe Browser MCP supports multiple agents simultaneously, offers 25+ tools, includes Google Workspace integration, a credential vault, sub-agent orchestration, and uses indexed markdown snapshots for lower token usage.

Can multiple AI agents control the browser at the same time?

Yes. Vibe MCP uses a relay daemon architecture where multiple AI clients (Claude, Cursor, VS Code, etc.) connect via WebSocket to a shared relay, which forwards requests to the Vibe extension. Each agent gets its own stdio MCP bridge.

Does Vibe Browser MCP require Chrome debug permissions?

No. Vibe Browser MCP uses content scripts and the Chrome Extensions API — no --remote-debugging-port or CDP required. Playwright MCP's extension mode also avoids debug ports, and it can manually attach via --cdp-endpoint if Chrome is already exposing CDP, but its default mode still launches a separate browser. Chrome DevTools MCP requires debug ports. Vibe works with your normal browser profile without any special launch flags.

What AI agents work with Vibe Browser MCP?

Vibe Browser MCP works with any MCP-compatible AI client including Claude Desktop, Cursor, VS Code (GitHub Copilot), OpenCode, Claude Code, Windsurf, Gemini CLI, and Codex.

How do indexed markdown snapshots work?

Instead of sending raw DOM or accessibility tree snapshots, Vibe captures indexed markdown snapshots with interactive elements labeled as [index:score]. Agents use these indices to click, fill, and interact with elements. This drastically reduces token usage and context pollution compared to full page snapshots.

Why does it matter that Vibe uses my real browser?

Playwright MCP and Chrome DevTools MCP launch a fresh, separate browser by default with no sessions, cookies, or extensions. Playwright MCP can reuse an existing browser through its extension flow or by pointing it at an explicit --cdp-endpoint, but that still requires extra setup and it still lacks multi-agent support. Vibe Browser MCP connects directly to the browser you are already using with zero configuration, so agents can interact with Gmail, Slack, GitHub, Jira, or any logged-in site without re-authenticating.

Real Browser MCP

Control Your Browser from
Any AI Agent|

Connect Claude, Cursor, VS Code, and more to your real browser — with all your sessions, cookies, and extensions intact. Works with Anthropic Claude Max, GitHub Copilot, Vibe AI, and BYOK providers. Multi-agent ready, internet-exposed relay, 25+ tools, and an open source MCP package for real browser control. Need the command-oriented OpenClaw flow? Use the dedicated Vibe Browser for OpenClaw page.

CLI

claude mcp add --transport stdio --scope user vibe -- npx -y --package @vibebrowser/mcp@latest vibebrowser-mcp

Or add to project-level .mcp.json

Requires Node.js and the Vibe Browser extension

Published MCP binaries: vibebrowser-mcp and vibe-mcp. For OpenClaw-style CLI flows, see the separate OpenClaw page.

25+ browser toolsMulti-agent relayInternet-exposed relayNo debug permissionsOpen source MCP package

Works with any MCP-compatible agent

Claude Code

OpenCode

Cursor

Claude Desktop

VS Code

Windsurf

Gemini CLI

Codex

Capability surfaces

What agents can actually do through Vibe Browser MCP

Package the browser, workspace, security, and coordination layers as explicit surfaces. That is the useful lesson from infrastructure products like Tavily, translated to a real-browser MCP.

Browser control

Drive the live browser session your operator already trusts

Open tabs, navigate, click, fill, drag, type, and scroll inside the same logged-in browser profile instead of a disposable automation runtime.

Why teams reach for this surface

This is the right surface when your agent needs real UI state, installed extensions, existing cookies, and human-visible execution.

Works on authenticated SaaS, portals, and internal tools

Keeps the human in the same browser context as the agent

Avoids extra browser launch flags or CDP ceremony

Representative tools

navigate_pageclickfill_formpress_keydrag

Packaging that technical buyers understand

Tavily makes its product legible by naming a small set of surfaces clearly. The equivalent move for Vibe is not “search” or “crawl”; it is browser control, snapshots, workspace actions, secrets, and agent coordination.

Proof points already present in this route

25+

published tools across browser, workspace, security, and agent orchestration

Local + remote

relay paths for laptop agents and internet-reachable runners

Multi-agent

one real browser session can be shared by multiple cooperating agents

Real sessions

cookies, tabs, extensions, and authenticated apps stay intact

How Vibe Browser MCP compares

Vibe Browser MCP is the only browser MCP that uses your real browser with all your logged-in sessions — no debug ports, no separate browser instance.

Feature	Vibe Browser MCP	Playwright MCP	DevTools MCP	Browser MCP
Uses your logged-in browser Playwright supports this via its extension mode, but defaults to launching a separate browser instance
No debug port required Playwright's default mode requires --remote-debugging-port; extension mode avoids this but adds token auth setup
Multi-agent support Multiple AI agents control the same browser simultaneously via relay daemon
Internet-exposed relay Expose your relay to the internet so remote agents can connect to your local browser from anywhere
Google Workspace integration Gmail search/send/draft/thread/message plus Calendar view/create/delete — 8 native tools
Credential vault Secure password manager that never exposes secrets to the LLM
Sub-agent orchestration Spawn sub-agents with isolated context and parallel tool execution
Standalone AI browser Also works as a standalone AI co-pilot directly in your browser
Open source Vibe's npm package (@vibebrowser/mcp) ships vibebrowser-mcp and the vibe-mcp compatibility alias as open-source binaries; the browser extension is not. BrowserMCP is similar
Telemetry to vendor Chrome DevTools MCP sends usage statistics and CrUX API calls to Google by default

Playwright MCP (Microsoft)

Launches a separate Playwright-managed browser. Requires CDP for existing browser connection. 27k GitHub stars.

Chrome DevTools MCP (Google)

Developer debugging tool using Puppeteer + CDP. Launches separate Chrome instance. Sends telemetry to Google by default. 25k GitHub stars.

Browser MCP (Namu)

Chrome extension-based like Vibe. Single agent only. Extension is closed-source. 5.8k GitHub stars.

Two ways to connect: local or remote

Run agents on your machine with the local relay, or connect any agent on the internet to your browser through our public relay server. Both support multiple agents simultaneously.

Local mode — agents on your machine

npx -y --package @vibebrowser/mcp@latest vibebrowser-mcp

  Claude Code          Cursor           OpenCode
       │                  │                 │
       ▼                  ▼                 ▼
   [vibebrowser-mcp]  [vibebrowser-mcp] [vibebrowser-mcp] ← stdio MCP bridges
       │                  │                 │
       └──────────────────┼─────────────────┘
                          │
                   ws://localhost:19888
                          │
                          ▼
                 ┌──────────────────┐
                 │   Local Relay    │  ← auto-spawned daemon
                 │  (localhost only) │
                 └──────────────────┘
                          │
                   ws://localhost:19889
                          │
                          ▼
                 ┌──────────────────┐
                 │  Vibe Extension  │  ← your real Chrome browser
                 └──────────────────┘

Remote mode
New
— agents anywhere on the internet

npx -y --package @vibebrowser/mcp@latest vibebrowser-mcp --remote <uuid>

  Cloud runner         Claude Code (laptop)     Cursor (office)
       │                      │                       │
       ▼                      ▼                       ▼
   [vibebrowser-mcp      [vibebrowser-mcp      [vibebrowser-mcp
    --remote <uuid>]         --remote <uuid>]         --remote <uuid>]
       │                      │                       │
       └──────────────────────┼───────────────────────┘
                              │
                    wss://relay.vibebrowser.app/<uuid>
                              │
                              ▼
                 ┌────────────────────────┐
                 │    Public Relay Server  │  ← hosted by Vibe
                 │  relay.vibebrowser.app  │     UUID-authenticated
                 └────────────────────────┘
                              │
                    wss://relay.vibebrowser.app
                              │
                              ▼
                 ┌──────────────────┐
                 │  Vibe Extension  │  ← your real Chrome browser
                 │   (at home)      │     connects on "Remote" toggle
                 └──────────────────┘

Remote mode setup

In the Vibe extension, go to Settings and enable MCP External Control in Remote mode

This connects your browser to relay.vibebrowser.app and generates a unique UUID

Copy your UUID from the extension settings page

Configure your agent with the --remote flag:

CLI

npx -y --package @vibebrowser/mcp@latest vibebrowser-mcp --remote YOUR_UUID

Share your UUID with any MCP-compatible agent on the internet. They connect to your browser through the relay — no port forwarding, no VPN, no firewall changes needed.

Multi-Agent

Multiple agents connect simultaneously — local or remote, no conflicts

Connect from Anywhere

Any agent on the internet can control your browser — just share your UUID

Authenticated

UUID + secret authentication — only you control who connects

25+ tools across 8 categories

Navigation, interaction, Google Workspace, credential management, skills, sub-agents, and more — all exposed via MCP.

Navigation

navigate_pageNavigate to a URL or go back / forward / reload

wait_forWait for text to appear on the page

Tab Management

list_pagesList open browser pages/tabs

new_pageOpen a new page with optional URL

close_pageClose a page by ID

Index-Based Interaction

clickClick by canonical ref like A7, M12, or 12

fillFill an input/select/contenteditable by canonical ref

fill_formFill multiple fields in one call

scroll_pageScroll page up or down

media_controlControl video/audio playback

Advanced Interaction

type_textType into the focused field

press_keySend keyboard keys and shortcuts

hoverHover to trigger tooltips/menus

dragDrag between elements or coordinates

Utility

waitWait N seconds

subagentSpawn sub-agent with isolated context

parallelExecute multiple tools in parallel

web_fetchHTTP fetch + text extraction

Inspection & State

take_snapshotComposite markdown + a11y + optional screenshot

take_md_snapshotTake a dedicated indexed markdown snapshot

take_a11y_snapshotTake a dedicated indexed accessibility-tree snapshot

take_html_snapshotTake a dedicated indexed HTML snapshot

take_screenshotJPEG screenshot with resize, grayscale, quality controls

evaluate_scriptRun JavaScript in the current page context

list_network_requestsList captured network requests

get_network_requestInspect one captured request/response

Google Workspace

gmail_searchSearch emails via Gmail API

gmail_get_messageGet full email content

gmail_get_threadGet all messages in thread

gmail_create_draftCreate email draft

gmail_send_messageSend email

calendar_viewView upcoming events

calendar_createCreate calendar event

calendar_deleteDelete a calendar event

Credential Management

secrets_managerList/read saved credentials

typein_secretFill form from vault (hidden from LLM)

settingsRead or update extension settings and skills

settings_listList stored settings keys and skill entries

memory_searchRecall stored context before acting

Skills Library

Turn workflows into reusable skills and share them across agents or teams.

Secrets Vault + Type-In

Internal password vault with a fill tool that never exposes secrets to the LLM.

Model & Agent Choice

Works with Vibe AI, Anthropic Claude Max, GitHub Copilot, and BYOK providers.

Set up in under 2 minutes

Install Extension

Install the Vibe AI Browser extension and enable MCP External Control in Settings.

Configure Your Agent

Add the MCP server config to your AI agent. Pick your agent below — copy, paste, done.

Configuration for every agent

Click an agent to see the config. Copy, paste, done.

Local mode defaults to relay port 19888 (no flag needed). The browser extension listens on 19889 — do not point the MCP server at the extension port.

CLI

claude mcp add --transport stdio --scope user vibe -- npx -y --package @vibebrowser/mcp@latest vibebrowser-mcp

Or add to project-level .mcp.json

Markdown-indexed pages, not raw DOM

Instead of dumping raw accessibility trees or DOM snapshots, Vibe extracts page content as clean markdown. Interactive elements get [index:score] labels that agents use to click, fill, and interact.

Lower token usage

Markdown is 3-5x smaller than raw DOM or accessibility trees

Scored interactions

Elements scored by relevance — agents focus on what matters

Automatic delivery

Page content sent before each agent turn — no tool call needed

page content (auto)

# @vibebrowser/mcp

Browser automation MCP and CLI package.

## Quick Start

[1:0.9] Run vibebrowser-mcp
[2:0.8] Run vibe-mcp
[3:0.7] View documentation

## Published binaries
- vibebrowser-mcp
- vibe-mcp (compatibility alias)

## Features
- Multi-agent relay
- 25+ browser tools
- Google Workspace integration

[4:0.6] Sign in
[5:0.5] Copy config
[6:0.3] Enable remote mode

Frequently asked questions

Ready to connect your AI agent to your browser?

Install the extension, add one config block, and start automating. Free, with an open source MCP package and browser CLI.

Questions? [email protected] · GitHub · npm

Personal automation Legal research automation LinkedIn recruiter automation Academic research automation

Control Your Browser from Any AI Agent|