Four tools with clear names and reasonable descriptions, but significant gaps in schema completeness and parameter documentation. All tools are READ_ONLY and well-scoped, but descriptions lack LLM-optimization guidance (e.g., when to use search_schemes vs get_scheme_details). Parameter descriptions exist but are sparse. Output schemas are not documented in the visible code. No error handling guidance or recovery paths visible.

• Document output schemas for all tools. For search_schemes, specify: [{scheme_name: string, field: string, answer: string, score: float}]. For get_scheme_details, specify the structure when field is omitted vs provided. For list_schemes, specify: {total_count: int, schemes: [string]}.
• Enhance parameter descriptions. For get_scheme_details.field, add: 'If omitted, returns all available fields (eligibility, benefits, description, application_process) as a single object.' For check_eligibility.user_context, clarify: 'Optional. When provided, returned alongside eligibility criteria so the agent can assess fit. Omit if no user context is available.'
• Add error handling guidance to tool descriptions. E.g., search_schemes: 'If no results are found, try a simpler query or call list_schemes() to discover available schemes.' get_scheme_details: 'If scheme_name is not found, call list_schemes() to see available options.'
• Clarify tool selection in descriptions. search_schemes: 'Use for semantic search when the user's intent is unclear or they ask open-ended questions.' get_scheme_details: 'Use when you know the exact scheme name and want specific fields (eligibility, benefits, etc.).'
• Add discovery guidance to list_schemes: 'Call this first to see all indexed schemes before using search_schemes or get_scheme_details. Returns total count and sorted list of scheme names.'