Glasses

Glasses MCP lets your AI agent capture accurate, device-specific screenshots of any website by connecting through a simple MCP server. It gives your assistant visual context, enabling it to understand layouts, designs, and dashboards just like a human would.

How to use

To use Glasses MCP with your MCP client, you connect the Glasses MCP server as a standard endpoint and then issue prompts that request screenshots. You can specify the URL, capture scope, format, and device emulation to fit your use case. Your AI will hand off the capture request to Glasses MCP and receive a structured result indicating success and the saved image path.

Practical usage patterns include taking a homepage screenshot, saving it in a preferred format, and testing how a site renders on different devices. You can also capture only the visible viewport for quick checks or opt for a full-page capture for thorough visual verification.

When you want to capture a site, you specify the target URL and optional parameters such as the output filename and device. Glasses MCP supports multiple devices that emulate common form factors, and you can choose between PNG or JPEG outputs.

How to install

Prerequisites: you need Node.js and npm installed on your system. Ensure you have a client that supports MCP integration and can load JSON configuration for MCP servers.

Step 1: Add Glasses MCP as an MCP server to your client configuration. Use the provided JSON snippet to register the server. This enables your client to run the Glasses MCP command on demand.

Step 2: Restart or reload your MCP client so it picks up the new server configuration.

Step 3: Run prompts to invoke the screenshot tool via Glasses MCP. The client will execute the underlying command and return the results.

Configuration snippet

{
  "mcpServers": {
    "glasses": {
      "command": "npx",
      "args": ["-y", "glasses-mcp"]
    }
  }
}

Tool reference: screenshot

The screenshot tool captures a website and saves the image to a local path. You can choose whether to capture the full page or just the visible viewport, and you can select the output format and device to emulate.

Supported devices

Device emulation lets you preview how a site looks on common form factors. Choose a device name that matches the target viewport when requesting a capture.

Error handling

If the capture fails, you receive a structured error message indicating the problem, such as an invalid URL or a site that could not be loaded.

Notes and tips

The tool does not log in to websites requiring authentication and may be affected by anti-bot measures on some sites. For dynamic pages, allow time for content to render before requesting a capture.

Development & contributing

Contributing follows standard MCP practices. Ensure dependencies are installed, build if required, and test the integration with your MCP client.

Available tools

screenshot

Captures a webpage as an image in png or jpeg format, with options for full-page or viewport capture and device emulation.