MCP server for semantic search over AWS Cloudscape Design System documentation
Server has two tools with clear naming and reasonable descriptions, but parameter documentation is inconsistent. Input schemas are properly defined with types. Tool descriptions are adequate but could be more explicit about prerequisites and return structures. No explicit error handling patterns documented. The tools follow a logical search-then-read workflow typical of documentation servers, but lack actionable error guidance and output structure documentation.
Read the FULL content of a documentation file. Use this tool SECOND, after finding the correct path via 'cloudscape_search_docs'.
Search the Cloudscape documentation index for relevant files. Use this tool FIRST to find the correct file paths. It returns a list of files with their relevance scores. It does NOT return the full content.
No documented output schemas for either tool. search_docs returns a formatted string listing files; read_doc returns wrapped markdown. LLMs cannot plan downstream steps without knowing return structure.
Error handling in read_doc is implemented (path validation, file checks) but NOT documented in tool description. LLM has no guidance on error cases or recovery steps. Errors return plain text (e.g., 'Error: File not found...') without structured error classification.
search_docs description mentions 'relevance scores' but output structure (title, path, filename, distance score) is not formally documented. LLM cannot reliably extract the returned path to pass to read_doc without parsing the formatted string output.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 63 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 36 | - | v1 |
Example values embedded in parameter descriptions ('e.g., "collection preferences", "table sorting props"' and 'e.g., "docs/components/button.md"'). LLMs frequently reuse examples literally rather than adapt to context, risking search/read failures.
No pagination or result limiting documented for search_docs. MAX_UNIQUE_RESULTS=5 is hardcoded in code but not mentioned in tool description. LLM has no visibility into why it received only 5 results or how to get more.