| name | eye |
| description | Analyze explicitly named local image files through the local Eye MCP visual observer. |
| argument-hint | "C:\path\image.png" [more image paths] <question> [--mode general|ui|document|compare] [--detail brief|standard|detailed] [--model name] |
| disable-model-invocation | true |
| allowed-tools | mcp__eye__analyze_images |
Eye
Use the local Eye MCP to analyze one or more image files named in this command.
User arguments:
$ARGUMENTS
Instructions
- Parse the arguments into one or more explicit absolute local image paths,
the user's question, and any optional
--mode, --detail, or --model
override.
- Do not try to open the images as native Claude Code image attachments. The
active main model may be text-only.
- Call
mcp__eye__analyze_images exactly once with the parsed paths,
question, and explicit overrides.
- Treat the returned report as observations made by a local vision model. You
must not claim that you directly saw, confirmed, or independently verified
facts that are present only in the images.
- For a purely visual question, answer with attribution such as "According to
the local vision analysis..." and preserve stated uncertainty.
- For a task involving code, logs, or text files, use the visual report as the
starting observation and separately verify any accessible non-visual
evidence needed for diagnosis or implementation.
- If the MCP tool fails, report the failure and recovery action without
inventing image contents.