MCP server providing canonical C++26 paper references and compiler-status lookup.
cpp26-ref has three well-named tools with clear descriptions and explicit input schemas. Tool names follow verb_noun patterns (lookup_paper, search, compiler_status). All parameters have type declarations and descriptions. However, output schemas are entirely undocumented, the code loads YAML files and returns markdown/structured data, but no schema is declared for LLMs to understand result structure. Error handling is minimal (server tolerates missing files but does not guide recovery). Parameter descriptions are adequate but lack format constraints and range specifications. The descriptions themselves are concise (50-120 chars) and explain the tool's purpose, meeting the 10-1024 character baseline. No dangerous defaults or security issues detected, all tools are read-only reference lookups.
Look up compiler support status for a C++26 paper.
Return the full markdown reference for a C++26 paper.
Free-text search over the C++26 paper index.
Output schemas are undocumented. The code returns markdown text for lookup_paper and structured data for search/compiler_status, but LLMs have no formal schema to understand what fields are present, their types, or how to extract data for downstream tools.
Parameter descriptions lack format constraints and examples. The 'paper_id' param says 'WG21 paper identifier (e.g. P2996, P2900)', this embeds example values that LLMs may reuse literally. Should state format constraint (e.g. 'Format: P followed by 4 digits').
The 'compiler' parameter in compiler_status is nullable but the description does not clearly state when to pass null or what the behavior difference is (return all compilers vs. a specific one). This dependency is undocumented.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
Error handling is minimal. The code tolerates missing files (search returns empty list, compiler_status returns 'unknown') but does not guide the LLM on what to do, e.g. 'No matching papers found. Try a broader search term or call lookup_paper with a known paper ID.'
The 'top_k' parameter default is 5 but has no description of why (signal vs. noise tradeoff, context limits). The description should explain when to increase top_k and what the cost/benefit is.