Sagaz toybox is a deliberately dangerous demo server with mixed annotation coverage. 12 tools with complete schemas (Zod-based, visible in source). Descriptions are present and clear (avg ~100 chars), but lack LLM-optimized guidance on when/why to use each tool. Tool names follow verb_noun convention well (list_*, get_*, create_*, delete_*, send_*, post_*, transfer_*). Critical gap: no output schemas documented, no pagination support on list tools, no error recovery guidance. Annotations deliberately mixed (some tools carry readOnlyHint/destructiveHint, others omitted for testing). Error handling wraps WorldError but provides minimal recovery hints. Composition is sound, each tool has one responsibility, but chaining IDs are not explicitly documented in responses.

• Document output schemas for all tools. E.g. list_contacts returns [{id: number, name: string, email: string, company: string|null}]. This enables LLMs to chain calls correctly.
• Add pagination to list_contacts, list_inbox, list_timeline, list_accounts: accept limit (1-100, default 20) and offset/cursor parameters; return total_count and has_more.
• Enhance descriptions with LLM-optimized guidance: 'List all CRM contacts. Call this first to discover available contacts before calling get_contact with a specific ID. Returns up to 100 contacts; use pagination for larger datasets.'
• Implement error recovery hints: 'Contact not found (ID 42). Try calling list_contacts to see available contacts, or create_contact to add a new one.'
• Add a confirm_delete or dry_run parameter to destructive tools (delete_contact, delete_tweet, transfer_funds) to prevent accidental data loss. Or wrap them in a separate confirmation tool.