Loading ToolBench
MCP server for generating images using OpenAI's DALL-E 3, DALL-E 2, gpt-image-1, gpt-image-1-mini, and gpt-image-2 models
The server defines 6 image generation tools with consistent, well-structured schemas across all tools. Tool names follow the verb_noun pattern (generate_image_*) and are sufficiently specific to distinguish between models. All tools have descriptions exceeding 20 characters and comprehensive input schemas with proper typing. However, no output schemas are documented, parameter descriptions lack depth in some cases (e.g., what 'auto' does for size/quality), and there is no error handling guidance or recovery patterns. The server demonstrates good baseline quality but lacks production-grade polish around output contracts and error scenarios.
Generate an image using OpenAI's DALL-E 2 model and save it to a file.
Generate an image using OpenAI's DALL-E 3 model and save it to a file. Supports high-quality image generation with style control.
Generate an image using OpenAI's gpt-image-1 model and save it to a file. Supports transparency, custom output formats, and high-quality generation.
Generate an image using OpenAI's gpt-image-2 model (released Apr 21, 2026). Supports flexible resolutions up to 3840px per edge, batch of up to 10 images, multilingual text rendering, and high-quality generation.
Generate an image using OpenAI's gpt-image-2 model with reasoning-aware "thinking" mode enabled. Thinking mode enables advanced reasoning for more complex or nuanced image generation requests.
No output schema documentation. Tools do not declare what they return (e.g., file paths, URLs, metadata). LLMs cannot plan downstream actions without knowing the response structure.
Enum values lack explanatory text. Parameters like 'size: auto' and 'quality: auto' are present but descriptions do not explain what 'auto' means or how the server selects values. This forces LLMs to guess behavior.
No error handling guidance. Tools do not specify what errors are retryable (e.g., rate limits, API timeouts) vs. user-fixable (invalid prompt content) vs. fatal. Error responses will not guide agent recovery.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 59 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 38 | - | v1 |
Generate an image using OpenAI's gpt-image-1-mini model and save it to a file. Cost-efficient alternative to gpt-image-1 with same features: transparency, custom output formats, and high-quality generation.
Parameter descriptions lack format constraints. For example, 'output' accepts a file path but does not specify: required directory must exist? Can paths be absolute or relative only? What happens on write failure? Ambiguity invites invalid inputs.
No dry-run or confirmation for write operations. All tools create files on disk (WRITE risk) but do not support preview or confirmation. Agents cannot safely test before committing.