PatentPincer defines 3 tools with clear, domain-specific descriptions (100-150 chars each). All tools have input schemas with typed parameters and descriptions. However, output schemas are not documented, the code shows SerpApi response handling but no explicit schema declaration for what agents receive. Parameter descriptions are good but lack format constraints (e.g., patent_id format, date format for 'before'). Error handling is present in code (status markers: ok, fixture, suppressed, error) but not surfaced in tool descriptions, leaving agents unaware of fixture vs. live evidence distinction. No tool annotations (readOnlyHint, idempotentHint) despite all three being read-only operations.

• Document output schemas for all three tools. Example for search_patents: 'Returns {status: "ok"|"fixture"|"suppressed"|"error", organic_results: [{title, link, snippet, patent_id}], total_results: number}'. Include status meanings.
• Add format constraints to patent_id and before parameters. Use descriptions: 'patent_id: Patent identifier (format: US-XXXXXXXX-B2, EP-XXXXXXX-A1, or similar). before: Priority date cutoff (format: priority:YYYYMMDD, e.g., priority:20230101).'
• Enhance tool descriptions with error guidance. Add: 'Note: Fixture results (status=fixture) are demo data only and cannot support actionable patentability verdicts. Only status=ok provides live evidence.'
• Add readOnlyHint=true to all three tools in MCP registration to signal they are safe for parallel execution and automatic retry.
• Clarify run_id purpose in parameter descriptions: 'Optional run identifier for spend-cycle grouping and ledger tracking. Omit for auto-assignment. Used to group related calls for cost analysis.'
• Add pagination guidance to search_patents and search_scholar if results can exceed 10-20 items. Document limit parameter and next_cursor if supported.