8 tools with complete JSON schemas and parameter descriptions, but significant gaps in tool-level descriptions and error handling. All tools have input schemas with proper types. Tool descriptions are present but brief (avg ~60 chars, below 194-char baseline). No output schemas documented. No error recovery guidance. No tool annotations (readOnlyHint/destructiveHint). Naming is clear and verb-based (list_, get_, read_, search_, send_, dump_). Parameters are well-typed with defaults and constraints (limits, pagination). However, descriptions lack LLM-optimization guidance (e.g., 'Получить публикации канала' is Russian and minimal; no 'when to use' or 'what it returns'). send_message and send_file lack confirmation/dry-run patterns despite being destructive. No documented output structure for any tool.

• Translate all tool descriptions to English and expand to 80-150 chars. Include 'when to use' and 'what it returns'. Example: 'list_channel_posts: Retrieve up to 100 posts from a channel, paginated backwards by timestamp. Use before read_messages to explore channel history. Returns: array of {id, text, timestamp, author_id}.'
• Document output schemas for all 8 tools. Specify returned fields, types, and which IDs are needed for chaining (e.g., send_message needs chat_id from list_chats or get_chat).
• Add confirmation_required parameter to send_message and send_file, or implement a separate confirm_send tool. Require explicit confirmation before executing destructive operations.
• Add tool annotations to FastMCP registration: mark list_*, get_*, read_*, search_* with readOnlyHint=true; mark send_* with destructiveHint=true. This enables LLM safety reasoning.
• Enhance error messages with recovery guidance. E.g., 'chat_id 12345 not found. Call list_chats() to discover available chats, or get_chat(id) to verify the ID.'