GazeSight by Medjio Teddy

GazeSight connects a Firefox tab you explicitly authorize to a local MCP (Model Context Protocol) runtime, so a compatible coding agent (such as Claude Code) can see and inspect the page you're actually working on — without launching a second browser.

Once you select a tab, your coding agent can:
- Capture the visible viewport (FAST, or STABLE which waits for the page to settle)
- Capture a specific region, element, or the full scrollable page
- Inspect the DOM structure, attributes and computed styles
- Compare two captures with a visual diff
- Read console errors, network failures and page status
- Preview the page at mobile/tablet/desktop widths in a temporary window GazeSight creates and manages

Optional features, off by default, each requiring your explicit permission:
- Continuous Access: keeps your selected tab authorized across same-origin reloads, navigation and responsive previews, so you don't have to reauthorize after every page change. Navigating to a different origin always requires a fresh authorization.
- Background Capture: lets your authorized tab stay capturable even when it isn't the active tab.
- Manual screenshot download: save a PNG of the current capture to your device.

GazeSight also provides short, local guidance to your coding agent's MCP connection (Agent Guidance, with an optional Visual Fidelity Mode for close visual-reproduction tasks) so it knows how to use these tools well; this never overrides your own instructions to the agent.

GazeSight does not operate a cloud capture backend. Captures and page data go from your browser to your local MCP runtime and then to the coding agent you configure — nothing is sent to a GazeSight server.