Loading ToolBench
ANATOMED MCP App connector — renders an interactive, region-isolated 3D anatomy widget inline in Claude.
Single tool with well-structured schema and comprehensive description. Tool name 'show_anatomy_region' uses verb_noun pattern appropriately. Input schema is complete with proper types and enums. Description is detailed (1,100+ chars) and explains WHEN to use the tool, what parameters control, and expected behavior. However, description is verbose and could be optimized for LLM parsing (baseline 50-200 chars). Output schema is not documented, LLMs cannot predict what the widget returns or how to chain subsequent calls. No error handling guidance (e.g., what happens with ambiguous queries, invalid structure names). Tool lacks destructiveHint/idempotentHint annotations despite being read-only.
Render an INTERACTIVE, rotatable 3D view of a specific anatomical REGION inline in the chat. Use this for any spatial / "show me" / "where is" / structure-relationship anatomy question. Pass one or more anatomical structures or a named region in `parts` (English, Latin, or Croatian) — e.g. ["cervical spine"], ["Femur", "Sciatic nerve"], ["hand bones"]. Named regions (cervical/thoracic/lumbar spine, hand bones, foot bones, carpus, tarsus, skull bones, neurocranium, viscerocranium) expand to all their members. Decomposed sternum and heart queries expand to bounded composites. Ambiguous queries return no structure plus explicit options; never treat an ambiguity as a successful render. Set `detail` to control surrounding context: "isolated" = only the requested structures (default; cleanest); "related" = also show the nearest neighbouring structures it passes through / runs near (translucent) — best for paths, routes, relations, "what does it pass", neurovascular bundles; "regional" = a wider surrounding context. The widget shows the focus structures solid + context translucent, with a legend the user can toggle per structure, plus rotate/zoom. Never the whole body.
Output schema not documented. LLMs cannot predict widget response structure or extract data for downstream operations.
Description exceeds 1,100 characters (baseline 50-200 for LLM optimization). Verbose prose wastes tokens and buries key details like 'ambiguous queries return no structure plus explicit options'.
No error handling guidance. Description states 'ambiguous queries return no structure plus explicit options' but does not explain how LLM should interpret or act on this response.
Tool lacks readOnlyHint annotation despite being read-only. Current spec (2026-07-28) supports tool annotations for safety classification.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2025-06-18+ | v2 |
Parameter 'detail' enum values lack descriptions. LLM must infer meaning from enum names alone; explicit descriptions for each enum value improve clarity.