Server exposes 2 tools with visible schemas and descriptions, but exhibits significant quality gaps. Tool naming lacks action verbs (diagnostator, bioSensor use non-standard conventions). Descriptions are present but generic and lack LLM-optimized context about WHEN to use each tool or what happens on failure. Parameter schemas are properly typed but descriptions are minimal. Output schemas are not documented, callers cannot see what fields bioSensor returns or how to chain results. The server relies on deprecated Sampling pattern (McpSchema.CreateMessageRequest within tool logic) rather than direct integration, indicating older spec alignment. No error handling guidance, no output field documentation, no idempotency hints. Tool composition is weak: diagnostator performs inline sampling (model calls within tool) rather than returning structured data for agent composition.

• Rename tools to action-verb format: 'diagnostator' → 'get_diagnosis' or 'diagnose_patient'; 'bioSensor' → 'get_heart_rate' or 'get_vital_signs'. This lets LLMs infer intent from the name.
• Document bioSensor output schema explicitly. Add a code comment or return type annotation showing: { pulse: number, state: string, sleepDeprivation: number } with ranges/enums. Include this in the tool description too.
• Remove inline sampling from diagnostator. Instead, return structured patient data (name, medical_profile, current_pulse) and let the agent or orchestrator decide whether to call an LLM. Tools should be pure data sources, not reasoning engines.
• Add range and default constraints to parameters. E.g. 'numberOfDays: integer, range 1 - 365, default 7' and 'name: string, 1 - 255 chars, full name or last name, unicode supported'.
• Add error handling descriptions. E.g. 'If patient name is blank or not found in system, returns 400 Bad Request with message. Agent should prompt user for valid name or call search_patients() to discover available names.'
• Add tool annotations via @McpTool: readOnlyHint=true for bioSensor (safe to retry), readOnlyHint=false or destructiveHint=true for diagnostator if it modifies patient records.
• Expand descriptions to 100 - 200 characters explaining when and why to call each tool. E.g. 'Get the current heart rate and sleep deprivation status for a patient. Call before diagnosing to include recent vital signs. Returns a map with pulse (bpm), state (normal|elevated|critical), and sleepDeprivation (0.0 - 1.0).'

Inferred effective spec: <=2025-11-25.

• Relies on Sampling (deprecated) - integrate directly with the LLM provider API