Summary
An agent that can look at your screen, repeatedly.
Multi-display enumeration and change-based keyframes are what separate it from a one-shot screenshot: you can show an agent what happened across ten minutes, not only what is on screen right now.
What it is
A Python MCP server for screen capture and recording. It enumerates connected displays, captures a full display or a region, resizes to a maximum width, and records change-based keyframes into a grid — including scheduled bursts of up to 10 minutes.
What you get
- Every connected display listed, with its position, size and which one is primary
- A full display captured, or a specific region of it
- Automatic resizing to a maximum width
- An optional delay before capture, or an instant one
- Screen activity recorded as change-based keyframes, output as a grid
- Scheduled burst capture, up to 10 minutes
- The most recent capture retrieved again
- No daily cap on captures
Requirements
Python, and pip install third-eye-mcp. Captures carry promotional messages in their metadata.
Setup effort
One command — pip install third-eye-mcp