Server has 6 tools with basic descriptions and input schemas. Naming is verb-forward (process_, search_, get_, list_, review_) and mostly clear. However, descriptions are terse (avg ~80 chars, below the 194-char baseline), parameter descriptions are minimal or absent, output schemas are undocumented, and error handling lacks recovery guidance. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). The review_resolve tool accepts free-form 'corrected_value' with no validation constraints. Missing pagination on search_documents despite returning lists. Overall: functional but below production baseline.

• Document output schemas for all tools. E.g., search_documents returns [{document_id: str, name: str, score: float, text_snippet: str}]. Include field types and meanings.
• Add parameter descriptions and constraints. E.g., 'corrected_value (string, 1-10000 chars): The human-verified corrected value for the field. Must not be empty.'
• Add pagination to search_documents: accept 'offset' (default 0) and 'limit' (default 5, max 50). Return {results: [...], total: int, offset: int, limit: int}.
• Enhance error messages with recovery hints. E.g., 'Unknown document_id "xyz". Call list_documents() to see available IDs.' for get_document_text.
• Add tool annotations via FastMCP: mark process_document and review_resolve with destructiveHint=true; mark search_documents, list_documents, get_document_text, review_queue_pending with readOnlyHint=true.