Server exposes 4 tools with complete input schemas and descriptions. Naming follows verb_noun convention (get_*, search_*, generate_*). Descriptions are present and substantive (100-250 chars), exceeding the 10-char minimum. However, output schemas are not documented, LLMs cannot predict return structure. Error handling is minimal; tools fail silently or return generic error dicts without recovery guidance. Parameters lack enums where appropriate (e.g., 'flow' accepts 'M'|'X' but is free-form string). No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles. generate_report is a WRITE operation but lacks confirmation/dry-run pattern.

• Add explicit output schema documentation for each tool. Example: 'Returns: {country: str, indicators: {code: [{year: int, value: float}]}}'. This enables LLM planning.
• Convert 'flow' parameter to enum: flow: Literal['M', 'X'] with description 'M for imports, X for exports'. Prevents hallucinated values.
• Add tool annotations: mark get_country_macro, get_trade_data, search_market_data with readOnlyHint=true; mark generate_report with destructiveHint=true.
• Implement error classification in API calls. Return {error: str, retryable: bool, guidance: str} instead of bare exception strings. Example: 'World Bank API timeout. Retryable. Try again in 30s.'
• Add pagination to get_trade_data: accept limit (default 50, max 500) and offset parameters. Return {rows: [...], count: int, total: int, next_offset: int|null}.