Arche Browser MCP Server

How to use

You connect your MCP client to either a remote HTTP endpoint or a local stdio server to start controlling the browser and PC. By using the core primitives you can automate web navigation, input, and screenshots, and you can execute shell commands or Python code to manage files, processes, and system tasks. Start in browser-only mode to validate web automation, then enable full PC control with local access when you need to run commands on the host machine.

Remote access is configured through an SSE server on the target machine. You obtain a connect URL that includes a token, then provide that URL to your MCP client. Your client uses that URL to issue commands securely over SSE.

To exercise the remote mode from a client, you configure an MCP server with the HTTP URL of the SSE endpoint. The example below shows how a client might point to the remote server using a token.

How to install

Prerequisites: you need Python 3.10+ and a compatible browser (Chrome, Chromium, or Edge) for browser automation. You also need an MCP client such as Claude Code to connect to the MCP server.

Install the Arche Browser MCP package from PyPI or install directly from the GitHub source, then run the MCP server as needed.

# 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

Configuration and usage notes

Browser automation is enabled by default when you run the main command. You can disable the browser portion to run only local PC control by passing the appropriate flag. You can enable remote access to control the host machine and browser from another device by starting the server with SSE transport and a port.

Key CLI options include enabling local control, disabling the browser, turning on SSE transport, and configuring authentication tokens. The system informs you when authentication is enabled and displays a token you must use to connect from remote clients.

Examples and practical patterns

Browser-only automation from a client: start the server in default mode and issue commands to navigate web pages, read texts, click elements, and capture screenshots.

Full PC control from a single client: start the server with local access to run shell commands, execute Python code, manage files, and perform system maintenance tasks.

Remote access with SSE: run the server on the host, retrieve the connect URL with its token, and configure your MCP client to connect to that URL for remote control.

Security and safety notes

Token authentication is enabled by default for remote SSE servers to prevent unauthorized access. Local PC control requires a deliberate opt-in using the local run flag. There are no sandbox restrictions on local control, so use these capabilities responsibly.

Troubleshooting tips

If you cannot connect remotely, verify that the SSE server is running on the host, check the port configuration, and confirm that your client is using the correct connect URL and token. If browser automation fails, ensure the browser is compatible and accessible by the MCP server, and check for any required permissions or cookies related to the automated session.

CLI and config references

{
  "mcpServers": {
    "remote": {
      "type": "http",
      "name": "remote",
      "url": "http://YOUR_IP:8080/sse?token=YOUR_TOKEN",
      "args": []
    },
    "arche": {
      "type": "stdio",
      "name": "arche",
      "command": "arche-browser",
      "args": ["--local"]
    }
  },
  "envVars": []
}

Tools and primitives overview

The MCP provides a focused set of primitives to cover web automation and local system control. Core primitives include shell, python_exec, and screen_capture for system tasks, and a broad set of browser tools for navigation, DOM interaction, and data capture.

shell

Execute shell commands from the host operating system using the full system shell (bash, cmd, PowerShell).

python_exec

Run Python code with access to the local filesystem, processes, and installed libraries.

screen_capture

Capture a screenshot of the desktop or active window.

file_read

Read contents of a file on the local machine.

file_write

Write contents to a file on the local machine.

file_list

List directory contents matching a pattern.

file_delete

Delete a file or directory.

file_copy

Copy files or directories.

file_move

Move or rename files or directories.

clipboard_get

Get current clipboard contents.

clipboard_set

Set the clipboard contents.

system_info

Retrieve OS, CPU, memory, and disk information.

process_list

List currently running processes.

process_kill

Terminate a running process by PID or name.

goto

Navigate the browser to a URL.

get_url

Get the current browser URL.

get_title

Get the current page title.

reload

Reload the current page.

go_back

Navigate back in the browser history.

go_forward

Navigate forward in the browser history.

get_text

Extract text from a DOM element identified by a selector.

get_html

Extract HTML from a DOM element identified by a selector.

click

Click a DOM element specified by a selector.

type_text

Type text into an input element.

select_option

Select a value from a dropdown element.

check_box

Check or uncheck a checkbox element.

scroll_to

Scroll the page to a specified position.

screenshot

Capture a screenshot of the current page.

pdf

Generate a PDF from the current page.

get_cookies

Retrieve cookies from the current browser context.

set_cookie

Set a cookie in the browser context.

storage_get

Read key from localStorage.

storage_set

Write key to localStorage.

evaluate

Execute JavaScript in the page context.

file Read

Read file contents for local workflows.

process_kill

Terminate a process by PID or name.