The server defines 3 tools with good naming conventions and descriptions. All tools are read-only and well-intentioned for domain specificity (ADNI documentation). However, there are notable gaps: (1) search_catalog and fetch_pdf lack explicit input schema definitions in the visible code, the @mcp.tool() decorator shows parameter names and descriptions but not formal JSON Schema with types; (2) list_categories has no input parameters but also lacks explicit schema declaration; (3) output schemas are not formally documented, responses are JSON strings without declared response structure; (4) error handling is minimal, no guidance for LLMs on recovery paths if a PDF fetch fails or a search returns no results; (5) parameter constraints (e.g., query length limits, URL format validation) are not specified. The tool naming follows good patterns (verb_noun: search, list, fetch), and descriptions are reasonably detailed (70-150 chars), but fall short of production-grade LLM-optimization. The server leans heavily on instructions embedded in FastMCP initialization rather than per-tool guidance.

• Add explicit JSON Schema to tool input definitions. Update tool registration to include 'schema' with 'type', 'properties', and required field lists. Example: search_catalog should declare {'type': 'object', 'properties': {'query': {'type': 'string', 'minLength': 1, 'maxLength': 200, 'description': '...'}}}.
• Document output schemas in tool descriptions or as a structured 'returns' field. Example: 'Returns a JSON object with keys: {query (string), total_matches (int), showing (int), documents (array of {title, description, url, category}), hint (string)}'.
• Add error handling with recovery guidance. Example: 'If no documents match the query, try shorter keywords or check list_categories() to see available document types. If fetch_pdf fails with a timeout, the PDF may be very large, try searching for a more specific term.'
• Constrain parameter values in descriptions. Add to search_catalog: 'query must be 1-200 characters, e.g., "MRI protocol", "data sharing agreement", "consent form". Avoid very long sentences.' Add to fetch_pdf: 'url must be a valid HTTP(S) URL from a search_catalog result.'
• Implement pagination for search_catalog. Add optional 'offset' (default 0) and 'limit' (default 20, max 100) parameters. Return 'total_matches' and 'has_more' flag so LLMs can iterate through large result sets.
• Add error classification to responses. When search returns no matches, return {'query': '...', 'total_matches': 0, 'error': 'No documents found. Try: 1) Different keywords, 2) Call list_categories() to explore available document types, 3) Search for "Alzheimer's" or "imaging"'}.