Loading ToolBench
Professional Scientific Image Analysis — AI-powered vision analysis tools specialized for scientific research images including microscopy, histology, western blot, gel electrophoresis, cell culture, chromatography, medical imaging, electron microscopy, scientific charts, and microarrays.
Three tools with complete input schemas and descriptions. Naming is clear and action-oriented (analyze_, ocr_, compare_). Descriptions are detailed (150-250 chars) and domain-specific for scientific image analysis. However, output schemas are not documented, LLMs cannot predict return structure. Parameter descriptions are in Chinese, reducing accessibility. No error handling guidance or recovery paths documented. Missing tool annotations (readOnlyHint). Validation is present (file existence, URL format, enum constraints) but error messages are not actionable for LLM self-correction.
Analyze scientific research images with domain-specific expertise. Supports multiple image types: general, microscopy, electron_microscopy, histology, western_blot, cell_culture, medical_imaging, chromatography, gel_electrophoresis, plot_chart, microarray. Can accept local file paths or URLs. Generates professional scientific analysis using specialized prompts or custom user prompts.
Compare two images and identify differences. Supports local file paths and URLs for both images. Can compare from multiple aspects: 全面 (comprehensive), 外观 (appearance), 内容 (content), 颜色 (color), 构图 (composition).
Extract text from images using OCR (Optical Character Recognition). Supports both local file paths and URLs. Returns extracted text in specified language.
Output schemas not documented. LLMs cannot predict return structure (fields, types, nesting). Forces agents to guess what data is available for downstream use.
Parameter descriptions in Chinese only. Reduces accessibility for non-Chinese-speaking developers and LLMs trained primarily on English. Best practice: provide English descriptions for international use.
No error recovery guidance. Validators reject invalid input (file not found, invalid URL, bad image_type) but error messages do not guide LLM on next steps. E.g., 'file not found' should suggest 'verify path or use image_url instead'.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
Tool annotations missing. No readOnlyHint, destructiveHint, or idempotentHint declared. All three tools are read-only (safe to retry), but this is not signaled to the MCP client.
Mutually exclusive parameters not documented. analyze_image and ocr_image accept both image_path and image_url; compare_images accepts both for each image. Descriptions should state 'provide exactly one of image_path or image_url' to prevent LLM confusion.