Argus MCP

A browser automation server for AI coding agents. Named after Argus Panoptes, the all-seeing giant of Greek mythology.

The Problem

AI coding agents are great at writing code, but they often can't see what they've built. When an agent creates a web page, implements a form, or fixes a UI bug, it has no way to actually open a browser and check if things work.

This creates a gap: the agent writes code blindly, hoping it works, unable to verify the results visually.

The Solution

Browser MCP gives AI agents the ability to:

• Open a browser and navigate to any URL
• Interact with pages - click buttons, fill forms, select options
• See the results - take screenshots and view what's on screen
• Verify their work - check if that button actually works, if that form submits correctly

It's like giving the agent eyes and hands for the web.

How It Works

The agent can:

1. Start a browser session
2. Navigate to a local development server or any website
3. Perform actions (click, type, scroll, etc.)
4. Take screenshots to see what's happening
5. Close the session when done

Screenshots are saved to files organized by session, making it easy to review what the agent saw and did.

Use Cases

• Testing a web app during development
• Verifying UI changes actually look right
• Filling out forms to test validation
• Checking that buttons and links work
• Debugging visual issues by seeing what the agent sees

Installation

Using npx (easiest)

No installation needed - just configure and run.

Global install

npm install -g argus-mcp

Note on Playwright

The first time you run the server, Playwright will need to download browser binaries. You can do this manually:

npx playwright install chromium

Configuration

Quick setup (recommended)

claude mcp add -s user argus npx argus-mcp

Manual configuration

Add to your Claude Code MCP settings (~/.claude/claude_desktop_config.json):

{
  "mcpServers": {
    "browser": {
      "command": "npx",
      "args": ["argus-mcp"]
    }
  }
}

Or if installed globally:

{
  "mcpServers": {
    "browser": {
      "command": "argus-mcp"
    }
  }
}

Available Tools

| Tool | Description | |------|-------------| | browser_start | Launch a new browser session | | browser_run | Execute actions (navigate, click, fill, etc.) | | browser_screenshot | Capture a screenshot of a specific region | | browser_close | Close a browser session |

License

MIT