Server provides 10 well-named tools with consistent verb_noun patterns (kanban_list_items, kanban_get_item, kanban_create_item, etc.). All tools have descriptions (avg ~80 chars, within baseline 34-392 range). Input schemas are present and properly typed with enums for constrained fields (status, item_type, priority). However, output schemas are completely undocumented, no tool describes what fields it returns, forcing LLMs to guess downstream field names. Error handling is absent: no recovery guidance, no actionable error messages, no distinction between retryable vs fatal failures. Parameter descriptions are minimal (e.g., 'Filter by assignee name' lacks format/type hints). Missing: pagination support for list tools, idempotency guidance for write operations, and confirmation patterns for destructive actions.

• Document output schemas for all tools. Example: kanban_get_item should return {item_id, title, status, priority, assignee, description, tags, created_at, updated_at}. This enables tool chaining and field extraction.
• Add pagination to kanban_list_items and kanban_get_board: accept limit (1-100, default 20) and offset parameters; return total_count and has_more. Cap results at 50 items per call.
• Implement error recovery guidance. Example: 'Item not found. Try kanban_list_items(status="in_progress") to find available items.' Categorize errors as retryable (network timeout) vs user-fixable (invalid status).
• Add dry-run support to write tools. kanban_move_item should accept dry_run=true to preview changes without committing. Prevents accidental state mutations.
• Expand parameter descriptions with format and constraint hints. Example: assignee should say 'Person name (string, case-insensitive). If unknown, call kanban_list_items() first to discover available assignees.' item_id should say 'Work item ID in format TYPE-NNN (e.g., FEAT-001, BUG-042)'.

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

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