Shelfmark has 5 tools with basic descriptions and minimal input schemas. Tool names follow verb_noun convention (corpus_stats, browse_folder, search_docs, get_file, search_emails), which is good. However, descriptions are terse (10-60 chars), lack context on WHEN to use each tool, and omit prerequisites. Most critically, input schemas are nearly empty: corpus_stats has {}, browse_folder has only 'prefix' with minimal description, search_docs/search_emails have only 'query'. No output schemas are documented. Error handling is absent, no guidance on what failures mean or how to recover. The server enforces governance (rights filtering, email sensitivity checks) but does not expose this in tool descriptions, leaving LLMs unaware of filtering behavior.

• Expand corpus_stats description to ~100 chars: 'Returns overall catalogue statistics: document count, total size, last refresh timestamp, and data freshness assessment. Call this first to understand corpus scope and recency.'
• Add output schema documentation for each tool. E.g., corpus_stats should document: {total_docs: int, total_size_bytes: int, last_refreshed: ISO8601, freshness_status: 'current'|'stale'}.
• Document pagination for search_docs and search_emails: add 'limit' and 'offset' parameters, return 'total_count' and 'has_more' in output, cap results at 50 per call.
• Expose governance constraints in tool descriptions. E.g., get_file: 'Returns full record including rights, authorship, and on-disk status. Records with rights=RESTRICTED are never returned. Email bodies are truncated unless explicitly requested.'
• Add error recovery hints to descriptions. E.g., search_docs: 'If no results, try broader keywords or use browse_folder to explore available facets. Returns empty list if no matches found.'