Critical gaps across all three tools. Tool definitions exist but lack proper schemas, parameter descriptions, and output documentation. No input validation schemas visible in source. Descriptions are present but minimal (10-50 chars). Tools are READ_ONLY but lack idempotent/readOnly annotations. No error handling guidance. This server cannot reliably guide LLM tool selection or parameter construction.

• Add explicit JSON Schema input definitions for each tool. searchMyKnowledge and searchPublicKnowledge need 'query' (string, required, 10-500 chars), 'limit' (integer, 1-50, default 20), 'offset' (integer, 0+, default 0). getDocument needs 'document_id' (string, required).
• Expand tool descriptions to 100-200 chars. Explain the distinction: searchMyKnowledge queries user's curated articles; searchPublicKnowledge queries public corpus. Include what each returns (e.g., 'Returns array of {id, title, excerpt, relevance_score}').
• Document output schemas. Define return type for each tool: array of search results with fields (id, title, excerpt, relevance_score, source) for searches; single document object with (id, title, content, metadata) for getDocument.
• Add parameter descriptions. 'query: Search terms (10-500 characters, required). Supports boolean operators (AND, OR, NOT) and phrase search ("exact phrase").' 'limit: Maximum results to return (1-50, default 20). Larger limits may increase latency.'
• Implement error handling with recovery guidance. E.g., 'No results found for query. Try: (1) broaden search terms, (2) remove filters, (3) call searchPublicKnowledge for broader corpus.' Include HTTP status codes and actionable messages.