visor-mcp provides 7 vision-focused tools with consistent naming (verb_noun pattern: ui_diff_check, extract_text_from_screenshot, etc.) and reasonable descriptions (80-120 chars each). However, critical gaps undermine production readiness: (1) Input schemas are visible but lack explicit type declarations and constraints, parameters are documented as strings with descriptions but no enum/pattern/minLength constraints; (2) No output schemas documented anywhere, LLMs cannot predict response structure; (3) No error handling guidance, tools return raw vision API responses without recovery hints; (4) Parameters accept URLs/paths/data-URIs but lack validation rules or format constraints; (5) No pagination, batching, or composition patterns despite image-heavy workloads. Tool names are clear and action-oriented, but parameter descriptions are minimal (10-30 chars) and lack format/constraint details required by the rubric baseline (72 chars average).
Analyzes charts, graphs, and dashboards
General-purpose image analysis tool for requests that do not fit a specialized tool
Diagnoses error messages, stack traces, and exception screenshots
Performs OCR on screenshots of code, terminals, configuration, and prose
Compares an expected/reference UI screenshot with an actual implementation
Converts a UI screenshot to artifact code representation
No output schemas documented. LLMs cannot predict response structure, field names, or types. Vision API responses are opaque to downstream tool composition.
Input parameters lack format constraints. 'image_source' accepts 'Data URL, HTTP URL, or absolute path' but no validation rules, size limits, or format patterns. LLMs may pass invalid URLs or oversized images.
No error handling guidance. Tools return raw vision API errors without recovery hints. LLMs cannot distinguish retryable failures (timeout, rate limit) from user-fixable ones (invalid image format).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | 2026-07-28+ | v2 |
Analyzes technical diagrams, architecture charts, and system designs
'analyze_image' is a generic fallback tool. Naming violates single-responsibility principle, LLMs cannot distinguish when to use it vs specialized tools (ui_diff_check, extract_text_from_screenshot, etc.). Ambiguous names cause wrong tool selection.
Parameter descriptions are minimal (10-30 chars). Rubric baseline is 72 chars. Missing format hints, constraints, and dependency documentation. E.g., 'prompt' param lacks guidance on instruction style or length.