Loading ToolBench
A FastAPI-based server for AI-powered fashion analysis using AWS Rekognition and OpenAI CLIP models. Processes clothing images to detect garments, extract semantic tags, and generate outfit recommendations.
This is an HTTP FastAPI server, not an MCP server. The codebase shows a REST API with clothing image tagging endpoints, but there is NO MCP protocol implementation visible. No tool registration via MCP standards, no MCP request/response handlers, no schema definitions in MCP format. The repository appears to be a standalone FastAPI application that would need to be wrapped in an MCP server layer to be usable as an MCP tool. Tool definitions are inferred from FastAPI route handlers rather than explicitly registered MCP tools. Parameter schemas exist in FastAPI but are not in MCP format. Descriptions are present but minimal (20-60 chars), below the 50-200 char optimum for LLM consumption. No error handling guidance for LLMs. No tool annotations. No pagination support visible despite tools accepting image uploads that could generate large results.
Upload an image to detect and tag multiple garments within a single image using AWS Rekognition and CLIP models.
Tag a clothing image with garment type and relevant features without storing the item.
Upload an image to get per-garment tags and attributes using AWS Rekognition for detection and CLIP for feature extraction.
Upload a clothing item image and return the predicted garment type and feature tags.
This is not an MCP server, it is a standalone FastAPI REST API. No MCP protocol layer detected. Tools are inferred from FastAPI routes, not registered as MCP tools with proper schemas.
Descriptions are extremely brief (20-60 chars), insufficient for LLM tool selection. 'Upload a clothing item image and return the predicted garment type and feature tags' lacks WHEN to call it, what the output structure is, and any prerequisites.
Tool name 'tag_image_endpoint' is not descriptive. Does not start with clear action verb and contains '_endpoint' suffix which is implementation detail. Should be 'tag_image' or 'analyze_garment_features'.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 33 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 32 | - | v1 |
No error handling or recovery guidance for LLMs. No input validation messages visible. Tools do not indicate what happens on invalid image formats, missing AWS credentials, or failed Rekognition calls.
No documented output schemas. LLMs cannot plan downstream calls without knowing response structure. What fields does 'tags' contain? Is it an object or array? What are 'garments'?
Parameter descriptions are minimal or absent. 'image_base64' lacks format guidance (size limits, supported formats). 'file' parameter lacks type clarity, are all image formats accepted? What is max file size?
'upload_clothing_item' has WRITE risk but description does not state that this modifies state or stores data. Agents need to know this is irreversible.
AWS Rekognition call may fail due to missing credentials or AWS service errors. No guidance provided to LLM on what to do when detection fails.
No pagination, result limits, or size constraints. Tools may return unbounded arrays of garments, exhausting context window. Response field 'recommendations' is undocumented.