Primo MCP has well-structured tool definitions with complete schemas and descriptions for all 5 tools. All tools follow verb_noun naming (primo_search, primo_get_record, primo_suggest, primo_cite, primo_export). Descriptions are substantive (100-250 chars) and explain WHAT each tool does and WHEN to use it. Input schemas are properly typed with enums for constrained fields (field, scope, sort_by, style, format). However, output schemas are not documented, responses are formatted strings rather than structured objects, forcing LLMs to parse unstructured text. Error handling returns plain text messages without recovery guidance or error classification. No tool annotations (readOnlyHint, idempotentHint) despite all tools being read-only. Parameter descriptions are clear but lack explicit format constraints (e.g., date_from/date_to state 'YYYY format' but no regex or validation rules visible in schema).

• Convert all tool responses from formatted strings to structured JSON objects with typed fields. E.g., primo_search should return {results: [{title, authors, year, identifiers, availability}], total_count, next_offset} instead of a formatted string.
• Add readOnlyHint: true to all 5 tool definitions in FastMCP registration to signal safety and enable agent optimizations.
• Enhance error messages with recovery guidance. Replace 'Error searching Primo: {e}' with 'Search failed: {error_detail}. Try: (1) simplify query, (2) check field/scope parameters, or (3) use primo_suggest to refine terms.'
• Add explicit format constraints to date parameters: date_from and date_to should include pattern: '^\d{4}$' in schema and minLength/maxLength in JSON Schema.
• Document output schema for each tool in the docstring. E.g., primo_search returns {results: array of {title: string, authors: array, year: int, identifiers: object}, total_count: int, next_offset: int}.