Three tools with mixed quality. 'add' and 'userInfo' have basic but adequate definitions; 'generateImage' has good parameter constraints but inconsistent description style. All tools have explicit schema registration via Zod. Parameter descriptions are present but some lack detail. No tool describes error cases or recovery guidance. Output schemas are implicit in the code (text or image returns) but not documented in descriptions.

• Add non-null descriptions to 'a' and 'b' parameters in 'add' tool: 'The first numeric operand' and 'The second numeric operand'.
• Expand 'add' description from vague to action-focused: 'Add two numbers together. Returns their sum as a text string.'
• Document 'generateImage' output schema in description: 'Returns a JPEG image (image/jpeg MIME type) generated from the prompt using the Flux model.'
• Add constraint detail to 'generateImage.prompt' description: 'A text description of the image to generate. Max 1000 characters; avoid special characters and URLs.'
• For error handling in 'generateImage', return structured error with retry guidance: 'Image generation failed: {reason}. {recovery_hint}' (e.g., 'Quota exceeded. Retry after 1 hour or try a shorter prompt.').
• Document 'userInfo' output schema explicitly: 'Returns an object with fields: userId (string), sessionId (string), email (string), firstName (string), lastName (string), imageUrl (string), role (string), metadata (object).'
• For role-gated 'generateImage' registration, pre-register the tool with a description noting it requires 'image_generation' role, so clients see it even if they cannot invoke it. Use permission error at invocation time.
• Add idempotentHint annotation to 'add' and 'userInfo' tools (read-only, safe to retry) and destructiveHint to 'generateImage' (stateful image generation, may consume quotas).