Strong schema coverage with Zod validation and consistent parameter descriptions. All 30 tools have explicit descriptions and input schemas with type constraints. Naming follows verb_noun convention (list_, get_, create_, update_, delete_, publish_, search_). However, descriptions are brief (avg ~50 chars) and lack actionable context for LLM selection. Output schemas are not documented, LLMs cannot predict response structure. Error handling returns raw JSON but lacks recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk stratification. Missing idempotency guarantees and confirmation patterns for destructive operations.

• Add output schema documentation to every tool. Define the structure of Post, Page, Category, Tag, Comment, and SearchResult objects in the tool definition (e.g., via outputSchema field or in description). This enables LLMs to predict response fields and chain tools correctly.
• Implement tool annotations: add readOnlyHint=true to all list_*, get_*, search_* tools; add destructiveHint=true to all delete_* and publish_all_drafts; add idempotentHint=true to get_* and list_* tools. This exposes risk metadata to the MCP protocol.
• Enhance error responses with recovery guidance. Instead of returning raw isError: true, return structured errors like {isError: true, message: 'Post not found (ID: 123)', suggestion: 'Try search_content() to find posts by title or author'}.
• Add confirmation/dry-run support for destructive operations. Implement a confirm_delete_post tool or add a dry_run parameter to delete_* tools. Alternatively, document that agents should call a separate confirmation tool before deletion.
• Expand tool descriptions to 100-150 chars with selection context. E.g., 'Get a single post by ID. Use this when you have the post ID; use search_content() to find posts by title or author.' This helps LLMs choose the right tool.

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

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