Arche Browser

MCP Server for Browser Automation and Full Local PC Control.

Control a real Chrome browser AND your entire computer from Claude Code or any MCP client.

Features

Full Browser Control: Navigation, clicks, typing, screenshots, and more
Full PC Control: Shell commands, Python execution, file system, clipboard, processes
Real Browser: Uses your actual Chrome with cookies, extensions, login sessions
Remote Access: Control browser/PC on any machine via SSE transport
Token Authentication: Secure remote access with auto-generated tokens
Minimal Design: Just a few powerful primitives that can do anything

Design Philosophy

Inspired by Eric Gamma's principles: Simple, Flexible, Powerful

Instead of hundreds of specific tools, Arche provides a few powerful primitives:

Primitive	What it does	What you can achieve
`shell()`	Execute any shell command	Volume, reboot, programs, services, ANYTHING
`python_exec()`	Execute Python code	Camera, Excel, AI, complex logic, ANYTHING
`screen_capture()`	Desktop screenshot	Visual feedback for AI

With just shell and python_exec, AI can literally control EVERYTHING on your computer.

Installation

# From PyPI
pip install arche-browser

# From GitHub
pip install git+https://github.com/GizAI/arche-browser.git

# One-liner (no install)
uvx arche-browser

Usage

Browser Only (Default)

arche-browser

Claude Code config:

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

Full PC Control

arche-browser --local

Claude Code config:

{"mcpServers": {"arche": {"command": "arche-browser", "args": ["--local"]}}}

PC Control Only (No Browser)

arche-browser --local --no-browser

Remote Access (SSE)

On the machine with Chrome:

arche-browser --sse --port 8080 --local

# Output:
# [*] Auth: ENABLED
# [*] Token: abc123...
# [*] Connect URL: http://localhost:8080/sse?token=abc123...

On Claude Code:

{"mcpServers": {"remote": {"url": "http://YOUR_IP:8080/sse?token=YOUR_TOKEN"}}}

Local Control Tools

Core Primitives

Tool	Description
`shell(command)`	Execute shell command (bash/cmd/powershell)
`python_exec(code)`	Execute Python code with full system access
`screen_capture(path)`	Capture desktop screenshot

Convenience Tools

Tool	Description
`file_read(path)`	Read file content
`file_write(path, content)`	Write file content
`file_list(path, pattern)`	List directory contents
`file_delete(path)`	Delete file or directory
`file_copy(src, dst)`	Copy file or directory
`file_move(src, dst)`	Move/rename file or directory
`clipboard_get()`	Get clipboard content
`clipboard_set(content)`	Set clipboard content
`system_info()`	Get OS, CPU, memory, disk info
`process_list()`	List running processes
`process_kill(pid/name)`	Kill a process

What You Can Do

# Volume control (Windows)
shell("powershell (Get-AudioDevice -Playback).SetMute($false)")

# Take a photo with webcam
python_exec("""
import cv2
cap = cv2.VideoCapture(0)
ret, frame = cap.read()
cv2.imwrite("photo.jpg", frame)
cap.release()
""")

# Create Excel spreadsheet
python_exec("""
import openpyxl
wb = openpyxl.Workbook()
ws = wb.active
ws['A1'] = 'Sales Report'
ws['A2'] = 1000
wb.save('report.xlsx')
""")

# System maintenance
shell("cleanmgr /d C:")  # Windows disk cleanup
shell("sudo apt autoremove")  # Linux cleanup

# Reboot computer
shell("shutdown /r /t 60")  # Windows
shell("sudo reboot")  # Linux

# Kill a program
process_kill(name="notepad.exe")

Browser Tools

Tool	Description
`goto(url)`	Navigate to URL
`get_url()`	Get current URL
`get_title()`	Get page title
`reload()`	Reload page
`go_back()`	Go back in history
`go_forward()`	Go forward in history

DOM & Input

Tool	Description
`get_text(selector)`	Get element text
`get_html(selector)`	Get element HTML
`click(selector)`	Click element
`type_text(selector, text)`	Type into input
`select_option(selector, value)`	Select dropdown
`check_box(selector, checked)`	Check/uncheck
`scroll_to(x, y)`	Scroll page

Screenshots & PDF

Tool	Description
`screenshot(path)`	Take screenshot
`pdf(path)`	Generate PDF

Cookies & Storage

Tool	Description
`get_cookies()`	Get cookies
`set_cookie(name, value)`	Set cookie
`storage_get(key)`	Get localStorage
`storage_set(key, value)`	Set localStorage

JavaScript

Tool	Description
`evaluate(script)`	Execute JS

CLI Options

arche-browser [OPTIONS]

Mode:
  --local          Enable full local PC control
  --no-browser     Disable browser tools (requires --local)

Transport:
  --sse            Run as SSE server for remote access
  --port PORT      SSE server port (default: 8080)

Browser:
  --headless       Run Chrome in headless mode

Authentication:
  --no-auth        Disable token authentication
  --token TOKEN    Use specific auth token
  --show-token     Show current auth token
  --reset-token    Generate new auth token

Architecture

arche_browser/
├── __init__.py    # Package exports
├── __main__.py    # CLI entry point
├── chrome.py      # Chrome process management
├── browser.py     # Full CDP browser automation
├── server.py      # MCP server with all tools
├── auth.py        # Token authentication
├── local.py       # Local PC control primitives
└── sites/
    └── chatgpt.py # ChatGPT-specific client

Security

Token Authentication: Remote SSE servers require authentication by default
Explicit Opt-in: Local PC control requires --local flag
No Sandbox: Local control has NO restrictions - use responsibly

Requirements

Python 3.10+
Chrome, Chromium, or Edge browser (for browser tools)
MCP client (Claude Code, etc.)

License

MIT