Model Context Protocol server that suggests recipes from groceries text or images using OpenAI.
The server exposes a single tool `suggest_recipes` with a comprehensive schema and detailed parameter descriptions. The tool name is action-oriented (verb_noun form) and the description is substantive (145 chars). However, the schema reveals several quality gaps: the output schema is not documented in the visible code, optional parameters lack clear defaults and dependency documentation, and error handling guidance is absent. The tool accepts both text and image inputs (imageUrl parameter with URI format), which is sophisticated, but the interaction model between these inputs is underdocumented. The parameter descriptions are present and mostly clear, but some lack actionability, e.g., 'extraNotes' is generic and doesn't explain what types of constraints are supported.
Suggest recipes based on available groceries, dietary preferences, cuisine preferences, and other constraints. Analyzes text descriptions or images of groceries and generates personalized recipe suggestions.
Output schema not documented in visible source code. The tool description does not explain what structure is returned, what fields to expect, or how the result should be used downstream. LLMs cannot plan follow-up actions without knowing the response shape.
Parameter 'imageUrl' and 'groceries' relationship is underdocumented. The description does not clarify: Can both be provided? Is one required if the other is absent? What happens if both are omitted? This creates ambiguity for LLM invocation.
Parameter 'vegetarian' description states 'When omitted the server will ask the user.' This implies a multi-turn interaction pattern, but the MCP protocol does not support server-initiated user prompts in this way. The tool either needs to have a default value, or the description must be rewritten to explain actual behavior.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 51 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 37 | - | v1 |
No error handling guidance. The tool description does not explain what errors might occur (e.g., invalid image URL, no recipes found, API quota exceeded, dietary constraints impossible to satisfy) or how the LLM should respond.
Parameter 'extraNotes' is vague. Description says 'Any other constraints or goals such as cooking time or equipment' but does not enumerate valid constraint types, expected format, or examples. LLMs will guess at content, producing inconsistent inputs.
No pagination or result limiting documented. If the tool returns many recipes, there is no indication of whether results are paginated, limited, or whether an LLM-provided limit parameter is honored. Large result sets can exhaust context windows.
All parameters are marked as optional (required: []). The tool does not declare what minimum inputs are needed to produce useful output. An LLM might call the tool with no parameters at all, which would fail.