Two tools with complete input schemas and descriptions. Naming follows verb_noun pattern (create-image, edit-image). Descriptions are adequate (100-150 chars) but lack WHEN/WHY guidance. Parameters are well-typed with enums and constraints. Critical gap: no output schema documented, LLMs cannot plan downstream operations. Error handling is minimal; no recovery guidance. File path validation is present but not described in parameter text. Missing tool annotations (destructiveHint for WRITE operations).

• Document output schema for both tools. Example: {success: boolean, file_paths: string[], images: {path: string, format: string, size_bytes: number}[], error?: string}. This enables agents to extract file paths for downstream use.
• Expand descriptions to 100-150 chars with WHEN guidance. E.g., 'Generate images from text prompts using OpenAI's gpt-image-1. Use when user requests image creation, artwork, or visual mockups. Returns base64-encoded images saved to disk.'
• Add error handling with recovery hints in the tool implementation. Catch OpenAI errors and return: {success: false, error: 'Rate limited. Retry in 60 seconds.', retryable: true} instead of raw exceptions.
• Add destructiveHint: true to both tool definitions (via server.tool() annotations if SDK supports, or document in description). Signals to agents that these operations consume API quota and write to disk.
• Clarify file_output parameter: 'Absolute path to save image (e.g., /home/user/image.png or C:\Users\user\image.png). If n > 1, index appended: image_1.png, image_2.png, etc.'