Three tools with clear verb-noun naming (search_companies, get_company, search_officers) and reasonable descriptions (120-180 chars). All tools have input schemas with typed properties and descriptions. However, output schemas are completely undocumented, LLMs cannot predict response structure. Parameter descriptions lack constraints (e.g., country code format, jurisdiction code examples). No error handling guidance. Missing pagination details despite API returning paginated results. Descriptions include example values (e.g., 'Tesla', 'Elon Musk'), which LLMs tend to reuse literally.

• Add explicit output schema to each tool definition. Document the structure returned by formatCompany() and formatOfficer() as JSON Schema objects in the tool's outputSchema field.
• Replace example values in descriptions with enum constraints. For country, add enum: ['us', 'gb', 'de', ...]. For jurisdiction, document the format (e.g., 'country_state' or 'country_code') and provide valid examples as a reference list, not in the description.
• Add pagination parameters (page, limit) to search_companies and search_officers. Document max results per page and total result count in the response. Cap default limit at 20-50 to prevent context window exhaustion.
• Implement error recovery guidance. Catch API errors and return structured messages: 'Company not found in jurisdiction us_de. Try search_companies() to find the correct jurisdiction code.' Categorize errors as retryable (rate limit, timeout) vs user-fixable (invalid jurisdiction).
• Add parameter validation hints to descriptions. E.g., country: 'ISO 2-letter country code (e.g., us, gb, de). Leave empty to search all jurisdictions.' jurisdiction: 'Format: country_state (e.g., us_de for Delaware) or country_code (e.g., gb for UK).'