Four tools with complete input schemas and descriptions. All tools are read-only with proper annotations. Naming follows verb_noun pattern (search_, get_, list_, build_). Descriptions are clear and action-oriented (avg 150 chars). Parameters have types and constraints (enums, min/max lengths). Output schemas are documented as text responses. Main gaps: output field documentation is implicit (returned as JSON text, not formal schema), no pagination support for search_prompts despite potentially large result sets, and error handling is minimal (only get_prompt returns isError flag).

• Add offset/limit or cursor-based pagination to search_prompts. Return a next_cursor field and document pagination in the tool description.
• Return structured objects instead of text-wrapped JSON. Define explicit output schemas with typed fields (id, slug, title, etc.) so LLMs can reliably extract data.
• Enhance error handling: categorize errors as retryable (API timeout) vs user-fixable (invalid slug). Return actionable messages like 'Prompt not found. Try search_prompts() with a keyword.'
• Add enum constraints or format hints to build_prompt_brief parameters. E.g., aspectRatio could be an enum: ['1:1', '16:9', '9:16', '4:3', '3:4'] instead of free-form string.
• Document the output schema for each tool in the description. E.g., 'Returns an object with fields: id (string), slug (string), title (string), prompt (string), media (array), sourceUrl (string).'