MCP server for DALL-E image generation via ChatGPT web interface
Single tool 'generate-image' has a reasonable schema with typed parameters and enums, but critical gaps undermine quality. The tool description (67 chars) is below the 194-char baseline and lacks context on when/why to use it. Parameter descriptions are present but minimal (10-30 chars each). No output schema is documented, LLMs cannot predict what fields the response contains. No error handling guidance; failures return generic error objects without recovery hints. The tool performs a destructive write operation (saves file) but has no confirmation/dry-run pattern. No tool annotations (destructiveHint) despite modifying the filesystem. Overall, this is a bare-minimum implementation that would confuse LLMs about expected outputs and error recovery.
Generate an image using DALL-E via ChatGPT web interface
No output schema documented. LLMs cannot predict response structure (success field, image URL, file path, dimensions, etc.). Forces guessing and downstream parsing errors.
Tool description is 67 characters, below 194-char baseline. Lacks WHEN to use it, prerequisites (ChatGPT login), and what it returns. LLMs cannot determine if this is the right tool for image tasks.
Destructive operation (writes file to disk) lacks confirmation/dry-run pattern and has no destructiveHint annotation. Agents may overwrite user files without warning.
Error responses are generic objects with 'error' and 'code' fields but no recovery guidance. E.g., 'VALIDATION_ERROR' tells LLM nothing about how to fix it or retry.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 37 | <=2025-11-25 | v2 |
Parameter descriptions are minimal (10 - 30 chars). 'File path where the generated image will be saved' is adequate, but 'The image generation prompt' lacks guidance on length, style, or constraints. Baseline is 72 chars per param.