Two well-named, read-only tools with complete input schemas and solid descriptions. search_reactions and get_reaction both follow verb_noun naming and have clear, actionable descriptions (180 - 220 chars). Input schemas are properly typed with required fields. However, output schemas are not documented, LLMs cannot predict the structure of returned objects. Error handling is minimal (generic catch-all). No parameter constraints (e.g., limit bounds are enforced in code but not declared in schema). Missing descriptions for response fields forces LLMs to infer structure.

• Add output schema documentation to both tools. Example for search_reactions: 'Returns {count: number, reactions: [{rhea_id: string, equation: string, ec: string[], status: string}]}'.
• Declare limit parameter constraints in inputSchema: add 'minimum': 1, 'maximum': 100, 'default': 15 to the limit property.
• Enhance error messages with recovery guidance. E.g., 'reaction not found' → 'Reaction ID not found. Try search_reactions() with a compound name or EC number.'
• Add pagination to search_reactions: accept 'offset' parameter and return 'total_count' and 'has_more' fields to enable multi-page queries.
• Clarify 'Keyless' in descriptions: 'No authentication required, this tool is free and publicly accessible.'