Server has 9 tools with complete Zod schemas converted to JSON Schema. Tool names follow verb_noun convention (create_draft, get_draft, search_drafts). Descriptions are present but brief (avg ~80 chars, below 194-char baseline). All parameters have type definitions and descriptions. However, output schemas are not documented, LLMs cannot predict response structure. Error handling is minimal; no recovery guidance. Missing tool annotations (readOnlyHint/destructiveHint). Composition is reasonable: tools are single-purpose and chainable via UUIDs. Security concern: no visible input validation or sanitization despite database access.

• Document output schemas for all tools. Example: create_draft returns {uuid: string, title: string, created_at: ISO8601}. This enables LLMs to chain tools and extract required IDs.
• Expand tool descriptions to 100-200 chars. Clarify WHEN to use each tool: 'search_drafts_db searches local database by text content; search_drafts opens the Drafts UI with filters. Use search_drafts_db for programmatic access, search_drafts for interactive browsing.'
• Add tool annotations via MCP SDK. Mark read-only tools (get_draft, search_drafts_db) with readOnlyHint=true. Mark destructive tools (create_draft, append_to_draft) with destructiveHint=true.
• Add input validation and error recovery. Validate action names against available Drafts actions; return 'Unknown action: X. Available actions: [list]' instead of a raw error. Validate folder enum values.
• Fix open_draft parameter relationship. Make one of uuid or title required, or document: 'Pass either uuid (exact match) or title (search by name). If both provided, uuid takes precedence.' Add validation to enforce this.