UniProtMCPServer has well-structured tool names (all verb-noun format: search_uniprot, get_protein_*) and complete input schemas with proper types. However, descriptions are generic and lack LLM-optimization guidance. Parameter descriptions are minimal (e.g., 'The UniProt identifier (e.g., P12345)' repeats across 4 tools). Output schemas are not documented, LLMs cannot predict response structure. Error handling returns raw HTTP status codes without recovery guidance. No parameter constraints (enums, ranges, patterns) despite domain-specific valid values. Tools lack idempotency hints and do not document when to use each variant (e.g., when to call get_protein_sequences vs get_protein_by_id). Baseline: avg tool description ~60 chars (well below 194-char production baseline); zero param descriptions beyond the name.

• Document output schemas for all tools. E.g., 'Returns: {results: [{uniprot_id: string, name: string, organism: string, ...}], total: number}'. LLMs need to know what fields to extract for downstream calls.
• Enhance error messages with recovery guidance. Replace 'Failed to fetch data: 404' with 'Protein not found. Verify the UniProt ID format (e.g., P12345) or try search_uniprot() with the protein name.'
• Add parameter constraints and format guidance. E.g., 'uniprot_id: UniProt accession (format: 1 letter + 5 alphanumeric, e.g., P12345). Invalid format returns 404.' Use enums for 'limit' if there are practical bounds.
• Consolidate or clarify the four get_protein_* tools. Add a 'fields' parameter to get_protein_by_id (e.g., fields=['sequence', 'functions', 'structures']) to reduce tool proliferation, or document when each is preferred in descriptions.
• Add pagination to search_uniprot. Accept 'offset' and 'limit', return 'total_count' and 'has_more' in response. Cap default limit at 20 and max at 100 to prevent context window exhaustion.