Server defines 8 tools with Zod schemas and descriptions in Russian. All tools have input schemas with type definitions and parameter descriptions. However, descriptions are brief (avg ~80 chars), lack actionable context for LLM selection, and omit error recovery guidance. No output schemas documented. Parameter constraints (enums, ranges) are minimal. Tool names follow verb_noun pattern correctly. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels. Error handling is absent from visible code.

• Document output schemas for all 8 tools. E.g., get_user should return {user_id: number, name: string, photo_url: string, ...}. Include this in tool descriptions or a separate schema reference.
• Expand tool descriptions to 80-150 chars. Add WHEN to use (vs similar tools), WHAT is returned, and any prerequisites. E.g., 'Get wall posts (owner_id required). Returns up to 100 posts with text, attachments, dates. Use search_posts for keyword filtering.'
• Add tool annotations: mark post_wall and send_message with destructiveHint=true; mark all get_* with readOnlyHint=true. This enables agents to reason about side effects and retry safety.
• Constrain 'fields' parameters with enums or regex patterns. E.g., get_user fields: enum ['photo_200', 'city', 'followers_count', 'screen_name', 'verified', 'status']. Document valid field names in VK API docs.
• Add error handling to tool handlers. Catch VK API errors (invalid user_id, permission denied, rate limit) and return actionable messages. E.g., 'User not found (user_id=999). Try search_users() with a name.' Include retryability classification.