paper-search-mcp has 4 tools with reasonable naming and acceptable descriptions, but exhibits significant gaps in parameter documentation and schema completeness. Tool names follow verb_noun convention (search_papers, download_paper, read_paper, get_available_sources), which is positive. However, input parameters largely lack type information in visible schema declarations, and output schemas are not documented. Three of four tools accept free-form string parameters (sources, paper_id, source) without enum constraints, despite sources having a known, finite set of 20+ values. Error handling is minimal, no guidance on recovery actions, no categorization of error types (retryable vs. fatal), and no actionable error messages visible. The 'sources' parameter in search_papers should be an enum rather than a comma-separated string, per pattern:constrained-input. Description quality is moderate (194-char baseline acceptable, but many descriptions lack specificity on output structure, prerequisites, and when to use). No tool descriptions explain what fields are returned or how results should be used downstream.

• Convert 'sources' parameter to enum: {type: 'string', enum: ['arxiv', 'pubmed', 'biorxiv', 'medrxiv', 'google_scholar', 'iacr', 'semantic', 'crossref', 'openalex', 'pmc', 'core', 'europepmc', 'dblp', 'openaire', 'citeseerx', 'doaj', 'base', 'zenodo', 'hal', 'ssrn', 'ieee', 'acm', 'all']} instead of free-form string.
• Add bounds to max_results: {type: 'integer', minimum: 1, maximum: 100, description: 'Maximum results per source (default: 10). Must be 1-100.'}
• Document return schema for search_papers: specify that it returns an array of objects with fields like {title, authors, year, url, paper_id, source, abstract} with type information.
• Document return schema for read_paper: specify text format (plaintext vs markdown), encoding, any pagination or truncation behavior, and approximate size ranges.
• Add error handling guidance to tool descriptions. E.g., 'If source API returns 429 (rate limit), retry after 30 seconds. If paper not found (404), try search_papers first to verify it exists.'
• For download_paper, document what the 'local path' return value contains, is it absolute or relative? Can it be used with read_paper? What if file already exists, overwrite or error?
• Add recovery suggestions for common failures in source parameter validation. E.g., if agent passes invalid source, respond: 'Unknown source: "invalid_source". Available sources: arxiv, pubmed, biorxiv, ... Call get_available_sources() to list all.'
• Clarify paper_id format per source. E.g., 'ArXiv: numeric ID like 2301.12345; PubMed: PMID like 12345678; BioRxiv: DOI like 10.1101/2023.01.01.001234'

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

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