glass MCP server

glass is a helper program that lets your AI actually open and use the app it is building for you. Instead of asking you "does this look right?", the AI can launch the app, click buttons, type text, and check the result on its own. It is handy if you or your team build apps with a visual interface and want the AI to test its own work.

What this MCP server does

You ask your AI to run and check an app. The AI tells glass to launch that app. glass takes a picture of the screen, reads the app's list of buttons and fields, and can click or type into them. It can also read the app's log messages and spot when the screen changes. Then the AI reports back what it saw or fixed.

Click to zoom

What you can do with it

• Launch a desktop, Android, or iOS app and see what is on screen
• Click buttons and type into text fields by their names
• Read the app's list of controls instead of guessing from a picture
• Take screenshots and compare two of them to spot changes
• Read the app's log messages while it runs
• Check that a change actually worked after fixing code

Try asking your AI

• “Use glass to run my app, click the Add button, and tell me if the new task shows up”
• “Launch the app, fill in the email field with test@example.com, and check the Save button works”
• “Take a screenshot before and after clicking Submit and tell me what changed”
• “Run the app with glass and show me the last 20 log lines”

What it gives back to you

You get answers in the chat: what the app looks like, which buttons and fields it found, whether a click worked, and short summaries of screen changes. Screenshots come back as images, and log lines come back as text. If the AI fixed code, it tells you what it changed and how it checked it.

Before you start

Who it's for

Developers and testers who build apps with a visual interface and want their AI to run and check those apps without doing every click by hand.