Single tool 'jellyfin_server' with a portmanteau design combining 9 distinct operations (status, info, health, logs, restart, shutdown, updates, tasks, transcode_queue) into one tool. Input schema is present with proper types and enums, but the tool violates single-responsibility principle. Description is adequate (95 chars) but generic. No output schema documented. Tool name does not start with action verb and masks multiple concerns behind an 'operation' enum. This pattern forces LLMs to reason about which operation to invoke rather than selecting a tool by name.

• Split 'jellyfin_server' into 9 separate tools: get_server_status, get_server_info, get_server_health, get_server_logs, restart_server, shutdown_server, get_server_updates, list_scheduled_tasks, run_scheduled_task, get_transcode_queue. Each tool name starts with action verb and has single responsibility.
• Document output schema for each tool. E.g., get_server_status returns {status: string, uptime_seconds: number, version: string}. This enables LLMs to chain operations and extract required fields.
• Add explicit parameter descriptions for conditional requirements. E.g., 'task_id (string, required only when operation=task_run): The ID of the scheduled task to execute.'
• Add error handling guidance to tool descriptions. E.g., 'If restart fails, check server logs via get_server_logs. If shutdown fails, verify no active streams via get_transcode_queue.'
• For destructive operations (restart, shutdown), add a confirmation pattern or dry-run option to prevent accidental execution by agents.

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

• Relies on Sampling (deprecated) - integrate directly with the LLM provider API