Server has 7 tools with basic naming (verb_noun pattern) and descriptions present, but quality is inconsistent. Tool descriptions are adequate (50-150 chars) but lack actionable context for LLM selection. Input schemas are visible and typed, but parameter descriptions are sparse or missing. Output schemas are not documented. Error handling exists but lacks recovery guidance. No tool annotations (readOnlyHint/destructiveHint). Composition is reasonable, tools chain together (create_room → send_message → poll), but idempotency and confirmation patterns are absent for destructive operations (battle_end_room).

• Document output schemas for all tools. E.g., battle_create_room returns {roomId: string, joinUrl: string, spectateUrl: string, participants: array, ...}. Include field types and descriptions.
• Expand parameter descriptions to 50-100 chars each. E.g., 'after' → 'Message ID to poll after (optional; for incremental updates, pass the last message ID received to avoid re-fetching old messages)'.
• Add tool annotations: mark battle_poll, battle_get_status, battle_list_rooms with readOnlyHint=true; mark battle_end_room, battle_interjection with destructiveHint=true; mark battle_send_message with idempotentHint=false (blocking call, not idempotent).
• Document error cases and recovery paths. E.g., 'If battle_send_message times out after 5 minutes, call battle_poll to check for new messages. If no response, the participant may have disconnected.'
• Add a dry-run parameter to battle_end_room (e.g., dry_run=true) to preview the conclusion without committing, or split into battle_preview_conclusion + battle_end_room.