Server has 5 tools with basic verb-noun naming (get_*, search_*, help). All tools have descriptions (10-200 chars, within baseline range). Input schemas are present with type declarations and parameter descriptions. However, output schemas are completely undocumented, no structured return types specified. Error handling is minimal: tools return plain strings with no guidance on recovery, retryability, or next steps. No parameter validation, no enums for constrained inputs (e.g., 'db' parameter accepts free-form strings). The 'help' tool is a workaround for missing discovery, indicates poor tool composition. Descriptions lack WHEN/WHY context needed for LLM selection. No pagination support despite search_ncbi returning up to 10 results. Overall: functional but below production baseline.

• Document output schemas for all sequence-fetching tools. E.g., get_nucleotide_sequence returns {sequence: string, accession: string, length: number, organism: string}. Use JSON Schema in tool metadata.
• Add enum constraints to 'db' parameters. Replace free-form string with enum: ['nucleotide', 'protein', 'gene', 'pubmed', 'taxonomy']. Prevents invalid API calls.
• Enhance error messages with recovery guidance. E.g., 'Accession NM_999999 not found. Try search_ncbi("gene name") to find valid accessions.' Categorize as retryable vs user-fixable.
• Add pagination to search_ncbi: accept 'limit' (1-100, default 10) and 'offset' parameters. Return {results: [...], total: number, next_offset: number|null}. Prevents context window overflow.
• Remove 'help' tool. Instead, write rich descriptions for each tool explaining WHEN to use it (e.g., 'Use this to fetch a known sequence by accession. If you only have a gene name, call search_ncbi first.').

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

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