The server provides 4 read-only tools with generally good naming and schema definition. Tool names follow verb-noun convention (search, get_*) and clearly indicate their action. All tools have non-empty descriptions (145-200 chars, within baseline 194 average). Input schemas are present and properly typed for all tools. However, there are notable gaps: (1) output schemas are not documented, callers cannot see what fields to expect from results; (2) parameter descriptions, while present, lack specificity about value constraints and ranges; (3) no error handling guidance is visible; (4) no pagination details despite search tools potentially returning many results. The codebase shows sophisticated filtering logic (_build_chromadb_filters) but this complexity is not adequately surfaced in tool descriptions.

• Add explicit output schema documentation for all tools. For search/search_topic, document the structure of each result object (fields: chunk_id, doc_id, doc_title, authors, year, page_num, section, chunk_type, score, text, metadata). For get_document, specify document structure (id, title, authors, year, full_text, chunks, metadata). For get_index_stats, list the stats returned (total_docs, total_chunks, chunk_types breakdown, year_range, collections, journals, etc.).
• Add constraint descriptions to all numeric and enum parameters. For 'limit' param: 'Maximum number of results (1-100, default 10)'. For year_min/year_max: 'Publication year (1900-2099)'. For 'chunk_types': specify valid values directly in description or link to schema enum.
• Document result limits and pagination. E.g. 'Returns up to 100 results. Use limit and offset parameters to page through large result sets.' If ChromaDB enforces a max internally, state it.
• Clarify the distinction between search and search_topic in their descriptions. E.g. search: 'Semantic similarity search with optional filters. Fast, direct vector matching.' search_topic: 'Topic-based search with oversampling (retrieves 3x results) and reranking to improve relevance. Slower but higher quality for specific topics.' This helps LLMs choose the right tool.
• Add error guidance to descriptions. E.g. 'If no results found, try broadening year_min/year_max or removing author/tag filters. If doc_id is invalid, use search() to discover valid document IDs.'
• Document that chunk_types, sections, and journal_quartiles parameters accept arrays (plural naming is clear, but confirm in description that multiple filters are AND-ed or OR-ed).

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry