Two well-named tools with comprehensive input schemas and clear descriptions. Tool names follow verb_noun convention (search_* pattern). Both tools declare their purposes clearly and include parameter constraints (enums, ranges). However, output schemas are not documented in the visible code, LLMs cannot see what fields to expect in search results. Parameter descriptions are good but could be more explicit about return format and use cases. The deprecated 'count' parameter is present alongside 'limit', adding complexity. No explicit error handling guidance is visible.

• Document the output schema for both tools. Specify that duckduckgo_web_search returns an array of objects with fields: {title: string, description: string, url: string, hostname: string}, and duckduckgo_news_search returns {title: string, excerpt: string, url: string, source: string, date: string, relativeTime: string}. Include this in tool descriptions or a separate output schema field.
• Remove or fully specify the 'count' parameter. Either delete it entirely (breaking change, requires major version bump) or document its exact precedence: 'If both count and limit are supplied, limit takes precedence.' Better: emit a deprecation warning in the response if count is used.
• Add an enum constraint for the 'region' parameter in duckduckgo_web_search. Extract VALID_REGIONS from the code and add it to inputSchema.properties.region.enum, so LLMs see the full list of valid options.
• Enrich tool descriptions with error handling guidance. Add: 'If the search returns no results, try: (1) relaxing time filters, (2) shortening the query, (3) removing special characters. If you hit a rate limit, wait 1 second before retrying.'
• Make result limit descriptions dynamic. Instead of hardcoding 'Returns max 3 results', state 'Returns up to [limit] results (default 3, configurable via DDG_MAX_RESULTS env var).' Or read the actual limit from config at initialization and bake it into the tool definition.
• Add optional output schema documentation to the tool registration. Most tools do not expose this in the protocol yet, but returning a comment in the tool description like 'Response format: [{ title, description, url, hostname }, ...]' helps LLMs understand the structure.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry