Two tools with complete input schemas and detailed descriptions. Tool names lack action verbs (contaminants_max_levels, contaminants_for_commodity are noun-heavy). Descriptions are strong (194 - 250 chars, well above baseline 34 chars minimum) and include use-case examples. Input schemas properly typed with required fields. However, NO output schemas are documented, LLMs cannot predict response structure. Parameters lack type constraints (enums, ranges). Error handling is basic (generic error text only). Security is adequate (read-only tools, no secrets). Composition is sound (two complementary tools, no overlap).

• Add outputSchema to both tools. Example: {type: 'object', properties: {provenance: {type: 'object', properties: {standard, edition, source, currency_warning, coverage}}, results: {type: 'array', items: {type: 'object', properties: {contaminant, commodity, maximum_level, unit, stated_as, source_page}}}}}
• Rename tools to action-verb form: 'get_contaminant_max_levels' and 'get_commodity_contaminants'. This helps LLMs parse intent from the name alone.
• Add enum constraints to 'contaminant' parameter. Extract valid values from cxs193.json and list them: ['lead', 'cadmium', 'arsenic', 'mercury', 'methylmercury', 'tin', 'aflatoxins', 'ochratoxin A', 'patulin', 'melamine', 'hydrocyanic acid'].
• Enhance error handling in src/server.ts. Catch specific error types and return structured errors with recovery hints. E.g., 'Contaminant "xyz" not found. Valid contaminants: lead, cadmium, arsenic, ... Try one of these or call get_commodity_contaminants() to explore by food.'
• Document the PROVENANCE object and result array structure in tool descriptions. State that every response includes edition, source, and currency_warning to prevent misuse of outdated limits.