Two tools with basic verb-noun naming (get_costs, get_resources). Both have descriptions and input schemas visible in server.py. However, schemas lack formal JSON Schema structure with explicit type declarations and constraints. Parameter descriptions are present but minimal (10-40 chars). Output is JSON strings, not structured objects. No error recovery guidance. Missing enum constraints for granularity and resource_types. tenant_id injection pattern is good, but no documentation of this security model in tool descriptions. No pagination support despite potential for large result sets.

• Add formal JSON Schema to tool inputs with explicit type, enum, minLength, and pattern constraints. Example: granularity should be enum: ['DAILY', 'MONTHLY', 'HOURLY'].
• Return structured Pydantic models or TypedDicts instead of JSON strings. FastMCP can serialize these automatically. This enables LLMs to understand response structure without parsing.
• Expand parameter descriptions to 50-100 chars with format hints. E.g., 'Start date in YYYY-MM-DD format (e.g., 2024-01-15). Defaults to 30 days ago.' instead of 'Start date (YYYY-MM-DD), defaults to 30 days ago'.
• Add error recovery guidance: 'If tenant_id is missing, ensure the agent is authenticated and has a valid session. Retry with a valid tenant_id.' instead of bare 'tenant_id required'.
• Implement pagination for get_resources with limit and offset parameters, and return a total_count field. S3 buckets and RDS instances can exceed 50 items, risking context window exhaustion.