Server has 4 tools with mixed quality. Tool naming follows verb_noun pattern (add, search_snp, get_articles, get_full_text_markdown) which is good. However, descriptions vary significantly in quality and completeness. The 'add' tool has a minimal 18-char description. Input schemas are present for all tools with proper types, but parameter descriptions are sparse or missing context. No output schemas are documented. Error handling is minimal, most tools return generic error strings without recovery guidance. The server mixes unrelated concerns (calculator + dbSNP + article retrieval) in one server, violating single-responsibility principle.

• Expand 'add' description to 50-100 chars explaining when to use it (e.g., 'Performs arithmetic addition of two integers. Use for simple math operations.').
• Document output schemas for all tools. E.g., search_snp returns {snp_id: string, grch38: {chr, pos, alleles, gene}, grch37: {chr, pos, alleles, gene}}. This enables LLM planning.
• Add per-item error tracking to get_articles and get_full_text_markdown. Return {doi: string, success: boolean, data?: {...}, error?: string} array instead of concatenated strings.
• Implement recovery guidance in error messages. E.g., 'SNP not found for query "rs999". Try search_snp("rs") to list available SNPs.' instead of generic 'Error searching for SNP'.
• Split server into domain-specific servers: mcp-calculator (add), mcp-dbsnp (search_snp, get_snp_info resource), mcp-pubmed (get_articles, get_full_text_markdown). Each server has one clear purpose.

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

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