Server exposes 10 tools with explicit schemas and descriptions in Chinese. All tools have InputSchema properties with type definitions and parameter descriptions. However, descriptions are brief (10-50 chars), parameter descriptions lack actionable constraints (ranges, formats, examples), and error handling is minimal. Tool naming uses verb_noun pattern (editor_addComponent, asset_create) but lacks clarity on mutual exclusivity (instanceID vs path parameters). No output schema documentation. Destructive tools (asset_deleteFolder, editor_deleteGameObject) lack confirmation patterns. Composition is reasonable, tools are single-responsibility, but parameter design forces LLM reasoning about ID vs path resolution.

• Add confirmation/dry-run support to destructive tools: asset_deleteFolder and editor_deleteGameObject should accept a 'confirm' boolean parameter or return a confirmation token requiring a second call.
• Expand tool descriptions to 100-200 chars. Include WHEN to use (e.g., 'Call after modifying scripts to check for compilation errors'), WHAT it returns (e.g., 'List of error objects with file, line, message'), and any prerequisites.
• Document output schemas explicitly. For console_getLogs, specify: returns {logs: [{index, level, message, timestamp}], total, hasMore}. For asset_create, specify: returns {path, guid, type}.
• Add parameter constraints to schema. Move 'ms: 100-120000' from description into schema minValue/maxValue. Add enum constraints for 'kind' in asset_create and 'level' in console_getLogs.
• Implement error classification. Return structured errors: {error: 'INVALID_PATH', message: '...', retryable: false, suggestion: 'Use Assets/ prefix'}. Guide LLM on next action.