The server has 8 well-named tools with mostly complete input schemas and descriptions. Tool names follow verb_noun convention (image_compress, content_latest, upload_to_s3, notify_task_complete). Input schemas are properly typed with enums, min/max constraints, and per-parameter descriptions. However, output schemas are completely undocumented, the code shows no indication of what these tools return, which is critical for agent composition and chaining. Error handling guidance is absent from descriptions. The content tools (content_latest, content_list, content_read, content_search) have good pagination/limits. Image tools have comprehensive format enums. S3 upload tool exposes metadata as object type but lacks guidance on expected structure. Descriptions are generally 30-100 characters (within baseline range), but lack WHEN to use each tool and what downstream operations become possible. No recovery guidance for failures.

• Document the return type of every tool. For content_latest, document: 'Returns: array of BlogEntry objects with fields {id: string, title: string, slug: string, summary: string, date: ISO8601, author?: string}'. For image_compress, document: 'Returns: {output_path: string, original_size_bytes: number, compressed_size_bytes: number, compression_ratio: number}'.
• Add recovery guidance to tool descriptions. Example for content_read: 'If the blog is not found by id or slug, try content_search() to locate it by title keyword.'
• Clarify mutually exclusive parameters. For content_read, add: 'Exactly one of id or slug must be provided. If both are given, id takes precedence.'
• Define the metadata object schema for upload_to_s3. State: 'metadata: object with string keys and string values (e.g., {"project": "myproject", "version": "1.0"}). These become S3 object user-defined metadata headers.'
• Distinguish image_compress vs image_convert. Add context: 'image_compress is best when you have a valid image and need to reduce file size. image_convert is best when you need to change formats (e.g., HEIF to JPEG for compatibility) and optionally compress.'
• Document the endpoint parameter for upload_to_s3: 'For S3-compatible services like MinIO or DigitalOcean Spaces, provide the endpoint URL (e.g., https://minio.example.com). For AWS S3, leave empty.'
• Add idempotency guidance: State which tools are safe to retry (all read tools: content_latest, content_list, content_read, content_search, image_compress, image_convert). Note upload_to_s3 and notify_task_complete may have side effects; recommend deduplication keys or conditional logic.