Eye2byte provides your AI agent with real-time visual context by capturing screenshots, audio, and screen recordings directly through the Model Context Protocol (MCP).
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install eye2byte
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install eye2byte using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is eye2byte?
Eye2byte is an open-source MCP server designed to bridge the gap between AI agents and the user's physical desktop environment. By providing eyes to your agent, it enables the processing of visual information such as UI layouts, terminal errors, and multi-monitor setups. This skill allows agents to analyze what the user is seeing, making it indispensable for debugging visual issues or following complex workflows that are difficult to describe in text. All data processing is focused on privacy, keeping captures stored locally on the user's machine with configurable auto-cleanup intervals.
As a core part of the Openclaw Skills ecosystem, eye2byte leverages the Model Context Protocol to provide structured Context Packs to the LLM. These packs include details about the environment, screen state, and even transcribed voice notes, ensuring the agent has a holistic understanding of the user's current situation without compromising security via external cloud storage.
eye2byte Use Cases
- Debugging visual bugs or layout issues that are difficult to explain via text.
- Capturing terminal output or error dialogs that cannot be easily copied.
- Providing voice-annotated screen captures to explain complex workflows to an AI agent.
- Monitoring long-running processes or animations through screen recording sequences.
- Recalling recent visual context without needing to perform a new capture.
How eye2byte Works
- The user or agent invokes an MCP tool like capture_and_summarize.
- Eye2byte captures the screen, window, or specific region based on the requested parameters.
- If audio is included, the tool records and transcribes the user's voice locally using Whisper.
- The captured media is processed and stored in a local directory on the user's machine.
- A structured Context Pack is generated, containing metadata about the environment, visible signals, and the likely situation.
- The resulting analysis is returned to the agent, providing it with high-fidelity visual context for its next task.
eye2byte Setup
To install eye2byte and integrate it with your Openclaw Skills environment, ensure you have Python and FFmpeg installed. Use the following command to install the package:
uv pip install eye2byte
For local use, the agent will automatically detect the server. For remote setups, run the MCP server with a secure token:
eye2byte-mcp --sse --token <your_secret_token>
Ensure the EYE2BYTE_MCP_TOKEN environment variable is configured in your openclaw.json file.
eye2byte Data Schema & Taxonomy
Eye2byte organizes its captures into a structured format stored locally. The data is categorized as follows:
| Data Type | Location | Description |
|---|---|---|
| Captures | ~/.eye2byte/output/ |
PNG screenshots and MP4 clips |
| Transcriptions | Local Cache | Text output from Whisper audio processing |
| Context Packs | JSON/Markdown | Structured summaries including Goal, Environment, and Signals |
| Configuration | openclaw.json |
Stores the MCP token and connection settings |
eye2byte Advanced Features
- Multi-monitor support allowing for full workspace stitching or specific display targeting.
- Timed delays for captures to allow the user to open menus or hover states.
- Integration with local Whisper for private, high-accuracy audio transcription.
- Configurable auto-cleanup to manage disk space by deleting old captures after a set number of days.
- Support for SSE remote transport, allowing agents on different machines to see the host screen via a secure token.
SKILL.md
Loading
Related Openclaw Skills
Security Check
A comprehensive security auditing and inspection tool designed to protect AI agent environments from malicious code and prompt injections.
WatchOrFight Rock Paper Scissors (RPS)
An on-chain Rock Paper Scissors arena on the Base network where AI agents stake USDC, play via commit-reveal rounds, and earn reputation.
SwitchBot OpenAPI Control
Integrate SwitchBot smart home devices into your AI workflows via the official OpenAPI v1.1.
Finally Offline Culture
An MCP-powered skill for browsing curated culture content and generating professional HTML newsletter digests on demand.
Tomoviee Text-to-Speech
A sophisticated AI voice synthesis skill that converts text into natural-sounding speech with customizable emotions and speed.
Text-Preserve Multilingual Image Rendering
An advanced AI agent skill that prevents text corruption in AI-generated images by pre-rendering non-Latin characters through a Canvas-based pipeline.
Featured*
