Loading ToolBench
Bringing Human Capabilities to Coding Agents - A Model Context Protocol server providing visual analysis, document processing, reasoning, and interaction capabilities
Scoring was not performed
mcp__reasoning__sequentialthinking has non-standard naming convention with 'mcp__' prefix and '__' separators. Violates verb_noun pattern and makes tool intent opaque. Should be named 'brain_sequential_thinking' or similar.
No output schemas documented for any tools. LLMs cannot plan downstream calls or extract response fields. Eyes tools likely return detailed analysis objects; brain tools return structured reasoning, but callers have no schema to work with.
mcp__reasoning__sequentialthinking has no visible input schema. Tool definition appears incomplete or inferred from name only.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 29 | <=2025-11-25 | v2 |
| 2026-03-09 | C | 61 | - | v1 |
brain_patterns_info description is vague ('Learning about available reasoning frameworks'), does not explain what call returns or how it fits agent workflows. Should clarify: 'Returns list of available reasoning patterns (SWOT, 5-why, cost-benefit) with descriptions and when to apply each.'
eyes_compare 'focus' parameter enum lists 4 options but description does not explain when to use each. LLM must guess whether 'differences' means visual diff, content diff, or both. Add: 'differences = highlight visual/content changes; similarities = shared elements; layout = structural comparison; content = text/data focus only'.
No error handling guidance visible. Tools lack descriptions of failure modes, recovery steps, or what to do if analysis fails. E.g., 'If image cannot be read, try uploading a higher-resolution version' or 'If document extraction fails due to encoding, check file format.'
brain_analyze_simple and brain_patterns_info have minimal parameter descriptions. 'Pattern-based reasoning' and 'Framework information' are too vague, they do not tell LLM what input is expected or what output structure to anticipate.
eyes_read_document 'pages' parameter accepts string like '1-5' but validation rules are not documented. What happens if user passes '1-1000' on a 10-page doc? Can 'pages' be a comma-separated list? Add format constraint: 'Range format: start-end (e.g., 1-5) or all. Must be valid page numbers.'
brain_reflect_enhanced 'originalAnalysis' parameter has minLength/maxLength constraints but no guidance on what constitutes 'analysis' vs other text. Is this meant for structured reasoning objects or freeform text? Add example: 'Accepts analysis blocks, findings summaries, or reasoning chains (50-5000 chars).'
No idempotency guarantees documented. Vision tools (eyes_*) are read-only, so safe to retry. But brain_reflect_enhanced and brain_analyze_simple descriptions do not clarify if they have side effects or are purely analytical. State explicitly: 'This tool is read-only and safe to retry.'