browser-rs-mcp: Lightweight Stealth Browser Server for AI Agents in Rust

Introduction

browser-rs-mcp is a lightweight, stealth-oriented Model Context Protocol (MCP) browser server built in Rust. It's designed to allow multiple AI agents to efficiently share a single, persistent Chrome instance. With over 64 Playwright-style tools, it provides robust capabilities for parallel web scraping, automation, and quality assurance, all without the need for a Node.js runtime. Each agent operates within its own isolated tab group, ensuring efficient resource utilization and clear separation of tasks.

Why Use and Key Benefits

browser-rs-mcp offers significant advantages, especially for multi-agent AI systems and web automation tasks:

Efficiency and Performance: Built in Rust, it compiles into a single, compact binary (~5.5 MB) with a minimal memory footprint (~6 MB RSS for the server), significantly less than Node.js-based alternatives.
Multi-Agent Collaboration: It allows multiple AI agents to share one logged-in Chrome browser and a persistent profile. Each agent controls only its own tabs, enabling parallel operations without each agent needing to launch its own browser instance.
Stealth-Oriented Design: The default headful mode uses a locally installed Chrome with a persistent profile and does not inject page patches, minimizing common automation signals. It includes reproducible detector runners for testing against bot detection.
Simplified Deployment: As a single Rust binary, it eliminates the complexities of Node.js runtimes and npm dependency trees, making deployment straightforward.
Comprehensive Toolset: It provides 68 browser_* tools, covering navigation, interaction, network requests, cookies, storage, and diagnostics, offering Playwright-style functionality.

Here's a comparison with Playwright/Puppeteer-based MCP solutions:

Feature	browser-rs-mcp	Playwright/Puppeteer-based MCP (Node)
Server runtime	Single Rust binary	Node.js + npm dependency tree
Release artifact	~5.5 MB	Runtime and packages installed separately
Server memory¹	~6 MB RSS	~180 MB RSS
Multi-agent control	One Chrome, owner-isolated tab groups	Separate coordination required
Browser control	Raw CDP over one multiplexed WebSocket	Playwright

¹ Historical maintainer measurements excluding Chrome, taken from idle local servers. Exact memory varies by OS, build, runtime, and workload; treat these figures as an order-of-magnitude comparison, not a benchmark guarantee.

Installation

Getting started with browser-rs-mcp is simple. A locally installed Google Chrome or Chromium browser is also required.

1. Install on macOS arm64 or Linux x64:

curl -fsSL https://raw.githubusercontent.com/maestrojeong/browser-rs-mcp/main/install.sh | sh
browser-rs --help

2. Install on Windows x64:

irm https://raw.githubusercontent.com/maestrojeong/browser-rs-mcp/main/install.ps1 | iex
browser-rs --help

To pin a specific release instead of following latest:

curl -fsSL https://raw.githubusercontent.com/maestrojeong/browser-rs-mcp/main/install.sh | AB_VERSION=v0.3.1 sh

Alternatively, you can build from source using Cargo:

cargo install --git https://github.com/maestrojeong/browser-rs-mcp ab-mcp

Set AB_CHROME if Chrome is not in a standard location.

Examples

Quick Start

After installation, you can run browser-rs and verify its operation:

1. Run the server:

browser-rs

2. Verify with an MCP client:

Point an MCP client at it and drive the browser. A headful Chrome window will open.

browser_navigate  ? https://example.com
browser_snapshot                          # returns the accessibility tree

The common workflow involves browser_snapshot, performing an action, and then inspecting the returned accessibility diff. Most interaction tools accept a snapshot ref or a CSS selector.

Connect an MCP Client

browser-rs speaks standard MCP over stdio. Any compliant client can connect with a simple configuration. For example, a generic JSON configuration for an MCP client would look like this:

{
  "mcpServers": {
    "browser-rs": {
      "command": "browser-rs"
    }
  }
}

For scenarios where several agents need to share one browser process and profile, you can use HTTP mode:

browser-rs --port 9321
# streamable HTTP: http://127.0.0.1:9321/mcp
# legacy SSE:      http://127.0.0.1:9321/sse

Configure your client with http://127.0.0.1:9321/mcp for streamable HTTP.

Multi-Agent Tabs

When using HTTP, each request can identify its owner (via an ?owner= query parameter or X-Browser-Owner header). New tabs are assigned to the request owner, and each agent can only list, switch, and control its own tabs. This allows multiple agents to share the same Chrome process and login state while maintaining task isolation.