Server provides two tools with explicit schemas and descriptions. Both tools have well-documented parameters with types and constraints. However, there are notable gaps in output schema documentation, error handling guidance, and parameter descriptions lack some actionable detail. The generate_image tool accepts file operations (save_to_file, filename) without documenting expected behavior on success/failure or output format. No per-tool error messages or recovery guidance. Naming follows verb_noun convention adequately (generate_image, list_models), though descriptions are somewhat generic. Parameter schemas are properly typed with enums for model selection.

• Document complete output schemas for both tools. For generate_image, specify: { image_url: string, base64_data?: string, file_path?: string, model_used: string, generation_time_ms: number }. For list_models, specify: { models: [{ id: string, name: string, provider: string, supports_batch: boolean, latency_ms: number }], total_count: number }.
• Add error handling guidance to tool descriptions. For generate_image: 'If API returns rate_limit error, wait 60 seconds before retrying. If invalid_prompt, simplify or shorten the prompt. If API key invalid, check OPENROUTER_API_KEY environment variable.'
• Enhance generate_image description to explicitly state file output behavior: 'When save_to_file=true, image is saved to ./output/{filename}.png. Returns file_path in response on success. On save failure, returns image data anyway so caller can handle manually.'
• Add constraints to prompt parameter: 'Text description (1-4000 characters). Model-specific limits may apply. See list_models for details. Longer prompts may increase latency 2-5x.'
• Document filename parameter format: 'Base filename without extension (alphanumeric, underscore, hyphen only). Max 255 characters. .png extension auto-appended. If file exists, error returned, use unique names or implement overwrite logic.'
• Add pagination/limit guidance to list_models description: 'Returns all available models. Typically 4-10 models. No pagination required. Each model entry includes id, display_name, provider, and estimated latency_ms.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry