Loading ToolBench
Catalog-domain MCP tools for Magebit_Mcp (read + write over products and categories)
Server defines 10 tools with consistent naming (catalog.resource.action pattern) and complete input schemas. All tools have descriptions and typed parameters with descriptions. However, output schemas are entirely undocumented, no response structure is defined anywhere in the codebase. Parameter descriptions are generic and lack actionable constraints (e.g., no enums for status/type_id, no ranges for numeric fields, no format specs). Error handling is absent from tool definitions. The test file confirms tool registration but provides no implementation details. Schemas are present but minimal; descriptions exist but lack LLM-optimization guidance.
Create a new category
Delete a category
Retrieve a single category by ID
List categories with optional filtering and pagination
Update an existing category
Create a new product
Delete a product
No output schemas documented. Tools return data but LLMs cannot predict response structure, field names, or types. Agents cannot plan downstream calls or extract required IDs for chaining.
Parameter descriptions lack actionable constraints. 'Product type (simple, configurable, etc.)' does not specify valid enum values. 'Product status (1=enabled, 2=disabled)' uses numeric codes without explaining why. LLMs cannot validate inputs or self-correct.
No error handling guidance in tool definitions. Agents do not know which errors are retryable, which require user input, or which are fatal. No recovery suggestions provided.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
Retrieve a single product by ID or SKU
List products with optional filtering and pagination
Update an existing product
Destructive operations (delete) lack confirmation or dry-run support. Agents can permanently delete products/categories without a safety gate or undo mechanism.
Pagination parameters (limit, offset) lack documented ranges. No maximum result count specified. Agents could request 10,000 products in one call, exhausting context or timing out.