This server has critical structural issues preventing production use. Transport is unknown/not visible in source. Tool definitions lack proper MCP registration patterns, no explicit schema objects with type/description pairs visible. Descriptions are present but generic (French-language, domain-specific jargon without LLM-friendly context). Parameters exist but lack formal JSON Schema typing. No error handling, validation, or recovery guidance. No tool annotations. This appears to be a Python CLI tool collection, not an MCP server.

• Implement MCP server wrapper: create a Python MCP server (using mcp library or similar) that exposes these CLI tools as MCP tools with proper JSON-RPC handlers and transport (HTTP+SSE or Streamable HTTP).
• Convert argparse definitions to formal JSON Schema. For each tool, define input schema with 'type', 'description', 'required', 'enum' (for constrained values like mode='light'|'dark'), and 'default' fields. Example: pace_target.pace should be {'type': 'string', 'description': 'Reference pace in min:sec per km (e.g., 4:30). Must be between 2:30 and 12:00.', 'pattern': '^\d{1,2}:\d{2}$'}.
• Rewrite descriptions in English for LLM clarity. Add WHEN/WHY context. Example: 'Adjusts a reference running pace for environmental conditions (temperature, humidity, wind, altitude, acclimatization). Use this to predict target pace for a specific date/location. Requires weather data from fetch_weather().'
• Add output schemas. Document what each tool returns: build_page returns {file_path, size_kb, warnings}; fetch_weather returns {csv_path, locations_count, date_range}; model returns {json_path, performance_index_range}.
• Implement error handling with recovery guidance. Replace sys.exit() with structured error responses: {'error': 'missing_file', 'message': 'activities_base.csv not found in ./data', 'recovery': 'Run fetch_weather() first to generate required CSV files', 'retryable': false}.