codeloop_visual_review
Run an LLM-driven visual inspection over the captured screenshots and return ranked issues (layout, contrast, alignment, missing elements).
When to use codeloop_visual_review
After capturing a screenshot, when you need an LLM-vision sanity check on the rendered UI (lined-up text, contrast, badge state, etc.).
Parameters
| Name | Type | Required | Description |
|---|---|---|---|
| run_id | string | no | Specific run to review. Defaults to the latest. |
| screen_name | string | no | Single-screen mode. |
| project_dir | string | no | Project root. |
What you get back
Per-screen findings with severity and pixel coordinates.
Typical prompts that trigger it
Cursor + Claude Code call codeloop_visual_review automatically when your prompt looks like one of these:
- “Compare the screenshot to the Figma frame.”
- “Visual review of the new dashboard page.”
FAQ
What does codeloop_visual_review do?
Run an LLM-driven visual inspection over the captured screenshots and return ranked issues (layout, contrast, alignment, missing elements).
When should an AI agent call codeloop_visual_review?
After capturing a screenshot, when you need an LLM-vision sanity check on the rendered UI (lined-up text, contrast, badge state, etc.).
What does codeloop_visual_review return?
Per-screen findings with severity and pixel coordinates.
Is codeloop_visual_review part of the CodeLoop MCP server?
Yes. CodeLoop ships as an MCP server installed via `npx codeloop init`. Cursor and Claude Code call it automatically once the User Rule (Cursor) or CLAUDE.md (Claude Code) is in place.