MCP server for querying AWS cost and usage data using natural language queries with MFA token support
Single tool 'get_aws_cost' has a reasonable name and input schema with three parameters (query, mfa_token, request_id). However, the description is generic and lacks actionable context for LLM selection. The schema defines parameter types and basic descriptions, but lacks critical details: no enum constraints on query patterns, no format specifications, no output schema documentation, and no error recovery guidance. The tool accepts free-form natural language queries without validation or examples of supported patterns, inviting hallucinated queries. MFA token handling is exposed as a parameter (security risk per pattern:secret-injection). No pagination, no result limits documented, no structured output schema provided. Error handling is minimal, no guidance on what to do if a query fails or MFA is required.
Get AWS cost information based on a natural language query.
MFA token exposed as tool parameter, security risk. Credentials must use server-side secret injection, not parameters. Agent traces log all parameters, leaking secrets into logs and prompt history.
No output schema documented. LLM cannot plan downstream operations or extract required fields. Tool returns natural language text via nlp_processor.generate_response_text(), unstructured and unparseable for agent composition.
Query parameter accepts free-form natural language with no constraints, enums, or examples of supported patterns. LLM will hallucinate unsupported query formats. No guidance on valid cost dimensions, time ranges, or grouping options.
Tool description (58 chars) is too generic: 'Get AWS cost information based on a natural language query.' Does not explain WHEN to use it, what it returns, or how it differs from other cost tools. Lacks actionable context for LLM selection.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 43 | 2026-07-28+ | v2 |
No error recovery guidance. If query parsing fails, MFA is required, or AWS API returns an error, the tool returns raw error text with no actionable next steps for the LLM. No categorization of retryable vs. fatal errors.