Image Generation MCP Server - Generate marketing creatives using Google Gemini and OpenAI Images
Two tools with complete input schemas using Zod validation and clear descriptions. Both tools follow verb_noun naming (image_gen_*) and include parameter constraints (minLength, maxLength, enums, URL format). Descriptions are marketing-focused and contextual (150-200 chars, within baseline). However, output schemas are not formally documented, responses are built ad-hoc via buildImageContent() helper, returning untyped content arrays. Error handling is minimal: only missing API key is caught; no guidance for API failures, rate limits, or invalid image URLs. No tool annotations (readOnlyHint, idempotentHint) despite both being read-only operations. Parameter descriptions include examples (e.g., 'Example: https://example.com/product.jpg'), which LLMs may reuse literally.
Generate high-quality marketing creatives using Google Gemini/Imagen models via OpenRouter. Best for photorealistic product scenes, lifestyle photography, and banner creatives. Optionally include a product image URL for product-integrated visuals.
Generate a high-quality marketing image using OpenAI's gpt-5.4-image-2 model via OpenRouter. Provide a prompt and optionally a product image URL for product-integrated visuals. Supports square, landscape, and portrait sizes.
Output schema not formally documented. buildImageContent() returns untyped content array with conditional image/text fields. LLMs cannot predict response structure for downstream planning.
No tool annotations. Both tools are read-only (generate images, no side effects) but lack readOnlyHint. Agents cannot distinguish safe tools from destructive ones.
Minimal error handling. Only OPENROUTER_API_KEY absence is caught. No guidance for API failures, invalid product URLs, rate limits, or malformed responses. Agents receive raw error messages with no recovery path.
Example values in parameter descriptions (e.g., 'Example: https://example.com/product.jpg'). LLMs tend to reuse examples literally rather than adapting to context. Use enum/format constraints instead.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | 2026-07-28+ | v2 |
No pagination or result limits documented. If OpenRouter returns multiple images or large base64 payloads, response could exhaust context window. No guidance on expected output size.