\r \r

Eye2byte — Screen Context for Your Agent\r

\r Eye2byte is an open-source MCP server (GitHub, PyPI) that lets you see the user's screen. Use these MCP tools only when the user explicitly asks you to look at something, debug a visual issue, or capture their screen.\r \r

Privacy & Data Storage\r

\r

• All data stays local. Captures are stored in ~/.eye2byte/output/ on the user's machine. Nothing is sent to external servers (except the vision model API the user configured).\r
• Auto-cleanup: Captures are deleted after N days (default: 7, configurable in settings). Set to 0 to disable.\r
• MCP token: When using SSE remote transport, the --token flag sets a bearer token stored only in the user's openclaw.json. Treat it like any API secret. The token is never logged or transmitted beyond the Authorization header.\r
• No telemetry. Eye2byte collects zero analytics or usage data.\r \r

Available MCP Tools\r

\r

capture_and_summarize\r

Screenshot the user's screen and get a structured analysis.\r \r Parameters:\r

• mode — "full" (default), "window", or "region"\r
• monitor — 0 = active monitor (default), 1/2/3 = specific monitor, -1 = ALL monitors at once\r
• delay — seconds to wait before capturing (useful for menus/tooltips)\r
• window_name — capture a specific app window by name (e.g., "chrome", "code")\r \r Use this when the user says things like "look at my screen", "what do you see", "debug this", or "what's wrong here".\r \r

capture_with_voice\r

Screenshot + voice recording + transcription. Returns both visual analysis and what the user said.\r \r Use when the user wants to describe something verbally while showing their screen.\r \r

record_clip_and_summarize\r

Record a short screen clip, extract keyframes, and analyze the sequence.\r \r Use when the user wants to show you something that changes over time (animations, workflows, step sequences).\r \r

summarize_screenshot\r

Analyze an existing image file. Pass a file path to get a structured analysis.\r \r

transcribe_audio\r

Local Whisper transcription of any audio file.\r \r

get_recent_context\r

Retrieve recent Context Pack summaries from previous captures.\r \r Use this to recall what you've seen recently without re-capturing.\r \r

What You Get Back\r

\r Every capture returns a structured Context Pack:\r \r

Goal           — what the user appears to be doing\r
Environment    — OS, editor, repo, branch, language\r
Screen State   — visible panels, files, terminal output\r
Signals        — verbatim errors, stack traces, warnings\r
Likely Situation — what's probably happening\r
Suggested Next Info — what you should ask or do next\r
```\r
\r
## When to Use Eye2byte\r
\r
- User mentions something visual ("this button is broken", "the layout is wrong")\r
- User asks you to "look at" or "check" something on their screen\r
- You need to see error dialogs, UI bugs, or terminal output the user can't easily copy\r
- User is debugging and visual context would help your diagnosis\r
- User asks you to monitor or watch something\r
- You want to verify your changes had the intended visual effect\r
\r
## Multi-Monitor Tips\r
\r
- `monitor=-1` captures ALL monitors stitched together — useful for seeing the full workspace\r
- `monitor=1`, `2`, `3` for targeting specific displays\r
- Default (`monitor=0`) captures whichever monitor has the active window\r
\r
## Setup\r
\r
Eye2byte must be running on the machine whose screen you want to capture:\r
\r
**Local (same machine):** Already configured if this skill loaded.\r
\r
**Remote (different machine):** The user runs `eye2byte-mcp --sse --token \x3Csecret>` on their local machine, and configures the MCP connection URL in openclaw.json.\r