Server defines 5 image generation/editing tools with complete input schemas and descriptions. Tool names follow verb_noun pattern (generate_image, edit_image, generate_story, generate_icon_set, generate_from_video). Descriptions are present and moderately detailed (100-200 chars typical). However, output schemas are not formally documented, responses are constructed ad-hoc in imageResult() helper without a declared return type schema. Parameters have types and descriptions, but some lack constraints (e.g., 'model' accepts free-form strings instead of enum). Error handling exists but is minimal, most validation happens inline without recovery guidance. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). generate_from_video is truncated in source, making its full schema unverifiable.

• Declare 'model' parameter as enum: {type: 'string', enum: ['nano', 'flash', 'pro']} to prevent hallucinated model names.
• Document output schema formally. Define return type as {type: 'object', properties: {content: {type: 'array', items: {type: 'object', properties: {type, text, ...}}}}} so LLMs know what fields to expect.
• Add tool annotations: all image generation tools should include 'destructiveHint: true' (they write files) and 'idempotentHint: false' (repeated calls with same prompt may produce different images).
• Enhance error messages with recovery guidance. E.g., 'Invalid model: got "gpt4", must be one of: nano, flash, pro' instead of throwing silently.
• Complete and verify generate_from_video schema in source. Currently truncated; full definition needed for evaluation.