Xenon is a Python agent harness, not an MCP server. Tool definitions exist but lack critical quality markers: descriptions are in Chinese and extremely brief (avg 8-12 chars), parameter descriptions are minimal or absent, no output schemas documented, and no error handling guidance. The codebase shows tool registration in tool_executor.py and tool_families/, but schemas are incomplete. Tool names lack clear verb-noun structure (e.g., 'code_index' vs 'search_code_symbols'). No evidence of LLM-optimized descriptions or parameter constraints. This is a coding agent framework, not a production-grade MCP tool server.

• Expand all descriptions to 50-200 chars in English. Example: 'code_index' → 'Search for code symbols (functions, classes, variables) by name or pattern. Returns file paths, line numbers, and symbol types. Use before refactoring to find all references.'
• Add parameter constraints: 'search_pattern' should document regex syntax, length limits (e.g., 1-100 chars), and examples. 'git_command' should list allowed values (status, diff, log, add, commit, branch) as an enum.
• Document output schemas for each tool. Example for code_index: {results: [{file_path: string, line_number: int, symbol_name: string, symbol_type: enum}], total_count: int}
• Rename tools to verb-noun form: 'code_index' → 'search_code_symbols', 'ast_analyze' → 'analyze_ast_structure', 'diff_preview' → 'preview_diff'.
• Add error handling and confirmation for destructive tools. 'refactor' and 'git' should support dry-run mode and return actionable error messages (e.g., 'Rename failed: symbol used in 5 files. Call code_index to find all references first.').