Server has 5 tools with clear verb-noun naming (telegram_whoami, telegram_list_chats, telegram_get_messages, telegram_search_messages, telegram_send_message). All tools have descriptions (avg 95 chars, within 10-1024 range). Input schemas use Zod with type constraints and defaults. However, parameter descriptions are sparse or missing: 'chat' param in telegram_get_messages lacks description; 'query' in telegram_search_messages has none; 'message' param has no description. Output schemas are not documented, responses are text-wrapped JSON without formal schema declarations. Error handling is minimal: no recovery guidance, no categorization of retryable vs fatal errors. Security: write allowlist is enforced via assertWritable(), but no audit logging visible. Composition is sound, tools are single-purpose and chainable (list_chats → get_messages → search_messages).

• Add descriptions to all parameters. For 'chat' param: 'Chat identifier: numeric ID, @username, or phone number'. For 'query': 'Search query (minimum 1 character, required)'. For 'message': 'Message text to send (required, minimum 1 character)'.
• Document output schemas formally. E.g., telegram_list_chats returns array of {id: string, name: string, kind: 'channel'|'group'|'dm', unread: number}. Include this in tool registration or response metadata.
• Add error recovery guidance. E.g., 'If chat not found, call telegram_list_chats() to see available chats and their IDs.' Categorize errors: retryable (network timeout), user-fixable (invalid chat ID), fatal (permission denied).
• Implement audit logging for telegram_send_message. Log caller identity, chat, message text, timestamp, and result to stderr or a structured log. Required for compliance.
• Add a dry-run mode to telegram_send_message via optional 'dryRun' boolean parameter. Return what would be sent without actually sending. Prevents accidental messages.