Harvard Dataverse MCP has basic tool definitions with schemas and descriptions, but falls short of production quality. All 4 tools have input schemas and descriptions, but descriptions are minimal (10-80 chars, well below the 194-char baseline). Parameter descriptions are sparse or missing entirely. No output schemas documented. Error handling is minimal. Tools lack the LLM-optimized descriptions and comprehensive parameter guidance needed for reliable agent selection and execution. The server reads from a public API with no state mutation, which is appropriate, but the interface design does not follow agentic tool patterns for clarity and composability.

• Expand tool descriptions to 100-200 chars. Example: 'search' → 'Search Harvard Dataverse for datasets, files, or collections by keyword. Returns up to 1000 results with pagination. Use this to discover datasets before fetching metadata with dataset() or dataset_files().'
• Add descriptions to all parameters. For 'type' in search: 'Filter results by type: dataset (default), file, or dataverse. Omit to search all types.' For 'sort': 'Sort by name (alphabetical) or date (newest first). Default: relevance.'
• Document output schemas for each tool. Example for search: 'Returns {data: [{id, name, type, url, description}], count_total, start}. Use id as persistent_id for dataset() calls.'
• Enhance error messages with recovery guidance. Example: 'Persistent ID not found. Verify format (doi:10.7910/DVN/...) or use search() to find valid IDs.' For 404: 'Dataset not found. Try search() with keywords from the title.'
• Add pagination guidance to search description: 'Results are paginated (default 25 per page, max 1000). Use start and per_page to iterate. Returns total count for planning.'
• Validate persistent_id format in dataset() and dataset_files() before calling API. Return actionable error: 'Invalid persistent_id format. Expected doi:10.7910/DVN/... or similar. Got: <value>'.