Loading ToolBench
Exposes ArtSleuth's computational art analysis capabilities as MCP tools, allowing AI assistants to perform art-historical analysis including style classification, brushstroke analysis, artist attribution, and forgery screening within a conversational context.
ArtSleuth defines 4 tools with complete JSON schemas and reasonable descriptions (avg 150 chars). All tools are READ_ONLY and follow verb_noun naming (analyze_, classify_, compare_, detect_). However, parameter descriptions lack actionable constraints (no enums, ranges, or format specs), output schemas are undocumented, and error handling guidance is absent. No tool annotations (readOnlyHint, etc.) despite being read-only. Descriptions are adequate but generic, they state WHAT the tool does but not WHEN to use it or what to expect in return.
Run the full ArtSleuth analysis pipeline on an artwork image. Returns style classification, brushstroke analysis, artist attribution, and optional forgery screening.
Classify an artwork by historical period, art-historical school, and subject genre.
Compare two artworks for stylistic similarity based on embedding-space proximity. Useful as a screening tool, but does not confirm shared authorship on its own.
Screen a painting for statistical anomalies relative to a reference artist's corpus, flagging potential forgeries.
Output schemas are completely undocumented. Tools return analysis results but LLMs cannot know what fields to expect (e.g., does analyze_artwork return confidence scores, brushstroke details, artist names?). This forces agents to guess and wastes tokens parsing unstructured responses.
Parameter descriptions lack actionable constraints. 'image_path' has no format spec (absolute vs relative? supported formats?). 'reference_artist' has no guidance on format (full name? surname only? case-sensitive?). Enums, ranges, and patterns are missing.
No error handling guidance. If image_path is invalid or reference_artist not found, what does the tool return? How should the LLM recover? No recovery hints or error classification provided.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | <=2025-11-25 | v2 |
Tool annotations missing. All 4 tools are read-only (no side effects), but readOnlyHint is not set in the schema. This prevents clients from optimizing caching and retry logic.
Descriptions are generic and lack WHEN/WHY guidance. 'Classify an artwork by historical period...' tells WHAT but not WHEN to call it vs analyze_artwork. No dependency hints or prerequisite guidance.