Server has 7 tools with consistent naming (verb_noun pattern) and basic schemas via Zod. All tools have descriptions and input schemas are visible. However, descriptions are generic (avg ~60 chars, well below the 194-char baseline), parameter descriptions lack detail on constraints/formats, output schemas are undocumented, and error handling is minimal. No tool annotations (readOnlyHint/destructiveHint). Lambda invoke accepts JSON as a string rather than structured object, forcing LLM to serialize. Most tools return raw JSON without pagination or result limits documented. Composition is reasonable (7 focused tools), but descriptions don't guide LLM selection or explain when to use each tool vs alternatives.

• Add output schema documentation to every tool. Example for ec2_list_instances: 'Returns array of {instanceId, state, type, publicIp, privateIp, launchTime, tags}. Max 100 instances; use filters to narrow results.'
• Expand descriptions to 100 - 200 chars. Example: 's3_list_buckets: List all S3 buckets in the account. Call this first to discover available buckets before using s3_get_object. Returns bucket names, creation dates, and region.'
• Add readOnlyHint: true to all read-only tools (ec2_list_instances, s3_list_buckets, s3_get_object, cloudwatch_get_metrics, iam_list_roles, sts_get_caller_identity) and destructiveHint: true to lambda_invoke.
• Change lambda_invoke 'payload' parameter from string to object type. Let LLM pass structured data directly: {type: 'object', properties: {...}, description: 'JSON object to pass to Lambda'}.
• Add limit and offset parameters to ec2_list_instances and iam_list_roles. Document max results (e.g., 'Returns up to 100 instances; use offset for pagination'). Return total_count in response.