Three tools with complete input schemas and descriptions. Naming follows verb_noun pattern (run_assessment, get_assessment_history, get_benchmarks). Descriptions are adequate (80-150 chars) but lack actionable context for LLM selection. Parameters are typed and constrained (enums for vertical/persona, integer ranges for limit). Output schemas are documented in code comments but not formally exposed in tool definitions. Error handling is basic, no recovery guidance or categorization. No tool annotations (readOnlyHint, idempotentHint). Missing dependency hints and downstream chaining IDs.

• Add readOnlyHint: true and idempotentHint: true to all three tool definitions in schemas.py. This signals to LLMs that these tools are safe to call repeatedly without side effects.
• Expand tool descriptions to 100-150 chars with explicit WHEN/WHY context: 'Retrieve past supply chain assessments for an organization. Call this after run_assessment to track improvement over time. Returns scores, timestamps, and narrative summaries.'
• Expose output schemas in the MCP tool definition (outputSchema field). Document that run_assessment returns {scores: {domain: number}, narrative: string, timestamp: string}, get_assessment_history returns {assessments: [{...}], total: number}, and get_benchmarks returns {benchmarks: {domain: [{metric, unit, world_class, industry_average}]}}.
• Add valid enum values to parameter descriptions: 'Industry vertical (required, one of: demand, procurement, manufacturing, inventory, logistics, warehousing, risk, sustainability)' instead of relying on schema alone.
• Implement error categorization in tool handlers. Return structured errors: {error: 'invalid_vertical', message: 'Vertical must be one of: ...', recovery: 'Call get_benchmarks() to see available domains'} instead of generic ToolError.