Single tool with adequate naming and basic schema, but significant gaps in parameter descriptions, output documentation, and error handling guidance. Tool description is good (203 chars, within baseline 194 avg), but parameter descriptions are minimal. No output schema documented. Error handling is present but not recovery-focused. Rate limiting exists but not described to LLM.

• Document the output schema explicitly. Return structured JSON with fields like {"results": [{"title", "description", "url"}], "total_returned": int, "query": string} instead of formatted markdown. This lets LLM parse results predictably and chain to downstream tools.
• Add JSON Schema constraints: set `"maxLength": 400` on query param, `"minimum": 1, "maximum": 20` on count param. Remove reliance on description text for validation.
• Enhance error messages with recovery guidance. Replace 'Rate limit exceeded' with 'Rate limit exceeded. Wait 1 second before retrying this query.' Replace 'Invalid arguments' with 'Invalid query: must be a non-empty string, max 400 characters.'
• Document actual DuckDuckGo API behavior in tool description. Clarify: 'Returns instant answer (if available) + up to N related topics. Total results may be fewer than requested if fewer are available. No pagination API supported.'
• Add input validation in code and return constraint violations as 400-style errors with the invalid value and constraint. E.g., if query.length > 400, return {isError: true, content: [{type: 'text', text: 'Query too long: {actual} chars. Max 400 allowed.'}]}.
• Consider whether markdown output is the right format. If downstream tools expect structured data, return JSON. Markdown is human-readable but wastes tokens for LLM parsing.