Server has 4 read-only AWS tools with basic descriptions and parameter schemas. Naming is clear and verb-based (list_*, get_*, describe_*). However, parameter descriptions are minimal or absent, output schemas are not documented, and error handling is generic. Tools return raw JSON strings rather than structured objects. No pagination support despite potential for large result sets (e.g., list_s3_buckets, list_ec2_instances). Security model is sound (read-only enforcement via _aws_call), but tool composition and output shaping need improvement.

• Document output schemas for all tools. Specify that list_ec2_instances returns an array of {InstanceId, State, Type, AZ} objects; list_s3_buckets returns {Buckets: [{Name, CreationDate}]}; get_cost_last_30_days returns {ResultsByTime: [{TimePeriod, Groups, Total}]}; describe_iam_user returns {User: {UserName, UserId, Arn, CreateDate}}.
• Add pagination to list_ec2_instances and list_s3_buckets. Accept optional 'max_results' (default 20, max 100) and 'next_token' parameters. Return a 'next_token' field if more results exist, and a 'total_count' field.
• Enhance error messages with recovery guidance. Instead of 'AWS error: {exc}', return 'Failed to list EC2 instances in region us-west-2: InvalidParameterValue. Verify the region name is valid (e.g., us-east-1, eu-west-1) and your AWS credentials have ec2:DescribeInstances permission.'
• Add region parameter validation. Define region as an enum of valid AWS regions (us-east-1, us-west-2, eu-west-1, etc.) or validate against boto3's available regions and return a helpful error if invalid.
• Return structured objects instead of JSON strings. Modify tools to return Python dicts/lists that FastMCP serializes, not pre-stringified JSON. This enables LLMs to reason about field types and compose downstream calls.