Server has 4 tools with complete input schemas and descriptions. Naming follows verb_noun convention (generate_image, get_job_status, list_jobs, get_estimates). Descriptions are present and adequate (50-150 chars), explaining what each tool does. However, parameter descriptions lack detail on constraints, formats, and dependencies. Output schemas are not documented, LLMs cannot infer what fields to expect from responses. Error handling is minimal; no recovery guidance provided. No tool annotations (readOnlyHint/destructiveHint) despite clear semantic differences between write (generate_image) and read-only tools.

• Document output schemas for all tools. Example for generate_image: 'Returns {job_id: string, status: "queued"|"processing"|"completed"|"failed", estimated_seconds: number}'. This enables agents to chain calls and extract needed fields.
• Expand parameter descriptions with constraints and rationale. For image_size, add: 'Output resolution: 0.5K=512×512, 1K=1024×1024, 2K=2048×2048, 4K=4096×4096. Larger sizes take longer and cost more.' For thinking_level, add: 'minimal=faster but lower quality; high=slower but more detailed.'
• Add tool annotations. Mark generate_image with destructiveHint=true (creates jobs, consumes API quota). Mark read-only tools with readOnlyHint=true. This enables agents to reason about side effects.
• Implement error recovery guidance. For generate_image, document: 'If status is "failed", check error field for reason (quota_exceeded, invalid_prompt, api_timeout). Quota errors are retryable after 60s; invalid_prompt errors require user to revise the prompt.'
• Add pagination to list_jobs. Accept optional limit (default 20, max 100) and offset parameters. Return {jobs: [...], total: number, has_more: boolean}. This prevents context window exhaustion as job history grows.