The Ensembl MCP server demonstrates solid definition quality with 10 well-structured tools covering a comprehensive genomics domain. All tools have explicit JSON Schema input definitions with type declarations, descriptions, and reasonable parameter constraints. Tool names follow the verb_noun pattern (ensembl_*) consistently. However, there are systematic gaps: (1) output schemas are completely undocumented, no guidance on what each tool returns or how to chain results; (2) parameter descriptions, while present, are verbose and could be more concise for LLM parsing; (3) error handling is minimal, no recovery guidance in descriptions; (4) some parameter relationships are underdocumented (e.g., how raw/page_size interact); (5) descriptions occasionally drift into implementation details rather than user intent. The baseline for production tools is 194 chars avg description, these are 200-350 chars, signaling over-explanation. All 10 tools are directly visible in src/handlers/tools.ts with explicit registration, so no inference penalty applies.

• Add outputSchema to all 10 tools documenting return structure, field types, and pagination shape (total_count, page, page_size, next_cursor where applicable). Without this, agents cannot chain results or extract IDs for downstream calls.
• Reduce all parameter descriptions to <100 characters. Move detailed format/constraint specs to JSON Schema validation (patterns, enums, minLength/maxLength) rather than prose.
• Remove implementation details (API endpoint paths, internal field names) from tool descriptions. Describe WHAT the tool does from the user's perspective, not HOW the backend works.
• Add 'If X fails, try Y' recovery guidance to each tool description. E.g., ensembl_feature_overlap: 'If no features overlap, try a larger region or different feature_types.' This guides agent error handling.
• Clarify parameter relationships explicitly. E.g., ensembl_regulatory: 'Provide exactly ONE of: region, protein_id, or binding_matrix_id. region overlaps regulatory features; protein_id finds regulatory elements affecting this protein; binding_matrix_id retrieves matrix details.'
• Rename 'ensembl_meta' → 'get_ensembl_metadata' or 'get_ensembl_info', 'ensembl_compara' → 'find_homologs' or 'get_gene_orthologs', 'ensembl_ontotax' → 'search_ontology' or 'find_biological_terms' to improve discoverability for non-genomics agents.
• Add maxItems constraints to batch array parameters (identifier, target_species) in JSON Schema. Enforce the stated 200-item limit formally rather than in prose.
• Document assembly parameter behavior more precisely. Currently says 'Only affects human queries' but doesn't explain what happens if a non-human species is queried with assembly='GRCh38'. Does it error, warn, or ignore?

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

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