MCP server that exposes AWS Bedrock Knowledge Base as MCP tools via Lambda + API Gateway + Cognito OAuth. Integrates with Snowflake Cortex Agent for cross-platform supply chain manufacturing intelligence.
Single tool with clear naming (retrieve_logistics_docs) and complete input schema. Description is adequate (194 chars) and explains the knowledge base domain well. However, output schema is undocumented, the tool returns formatted text but LLMs cannot see the structure of results (score, source, content fields). Parameter descriptions are present but minimal. No error recovery guidance. Lacks tool annotations (readOnlyHint, idempotentHint). Missing documentation of pagination behavior and result limits.
Search the supply chain logistics knowledge base for information about freight costs, shipping rates, customer returns, warranty claims, compliance documents, ISO certifications, customs duties, and material safety data. Use this tool when asked about logistics costs, shipping, returns, or regulatory compliance.
Output schema not documented. Tool returns formatted text with embedded metadata (score, source) but LLMs cannot parse structure. Should return structured JSON with result array containing {content, source, score} objects.
No tool annotations. Tool is read-only and idempotent but lacks readOnlyHint and idempotentHint in schema. Agents cannot infer safety properties.
Pagination and result limits not documented. numberOfResults capped at 10 in code but description says '1-10, default 5' without explaining what happens if more results exist or how to paginate.
Error handling lacks recovery guidance. Exception handler returns generic 'Knowledge base query failed' message. Should guide LLM: 'Query may be too broad. Try more specific terms or check KB_ID configuration.'
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
Parameter 'numberOfResults' description lacks constraints. Should state: 'Integer between 1 and 10 (default 5). Higher values return more results but increase latency.'