Three tools with complete schemas and descriptions, but descriptions are generic and lack LLM-optimized guidance. All tools have proper input schemas with typed parameters and descriptions. However, descriptions lack actionable context (when to use, prerequisites, what happens). Output schema (AgePredictionResult) is documented but only for predict_age; other tools' outputs are inferred. No error handling guidance, no recovery paths. Tool naming is clear (verb_noun pattern) but generic. Parameters are well-typed but lack constraints (enums, ranges). No tool annotations (readOnlyHint present in metadata but not in schema). Missing pagination, batch operations, and composition guidance.

• Expand tool descriptions to 100-200 chars. Add WHEN to use (e.g., 'Use predict_age_with_metadata when you have donor metadata; use predict_age for gene expression only'), prerequisites ('Requires space-separated aging-related gene names'), and what happens ('Returns predicted age in years and model confidence').
• Document output schemas for all three tools. Specify which fields are returned, their types, and what they mean. E.g., insilico_knockout should document that it returns {original_age, knockout_age, delta, gene_found_warning}.
• Add error handling guidance. Document failure modes: 'If gene_symbol not found, returns warning in response. If vLLM service unavailable, returns timeout error, retry after 30s. If gene_sentence malformed, returns validation error with expected format.'
• Convert free-form string parameters to enums. sex: enum ['male', 'female', 'unknown'], tissue: enum ['blood', 'brain', 'liver', 'skin', ...], smoking_status: enum [0, 1] with descriptions.
• Add tool annotations to schema. Mark all three tools with readOnlyHint=true in the schema definition so LLMs know these are safe, non-destructive operations.

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

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