Single tool (WhoAmI) with solid fundamentals. Tool is properly registered with @McpServerTool annotation, has a clear description, and explicitly declares metadata (ReadOnly=true, Idempotent=true, OpenWorld=false). Input schema is empty object (no parameters required), which is correct for this use case. Output is text-based (StringBuilder) rather than structured JSON, which limits downstream composability. Audit logging infrastructure is well-designed with proper claim filtering (allowlist pattern) and sensitive-data protection. However, the response format (unstructured markdown-like text) prevents LLMs from extracting fields for programmatic use. Tool follows security best practices by never logging argument values and filtering claims explicitly.

• Refactor WhoAmI output to return a structured JSON object: {"name": string, "email": string, "oid": string, "upn": string, "tenant_id": string, "token_expires_at": ISO8601 string}. This enables LLMs to extract and chain these fields into downstream tool calls.
• Rename tool from 'WhoAmI' to 'get_current_user' or 'get_session_context' to follow verb_noun convention. This improves LLM discoverability and reduces confusion with similar identity tools.
• Add explicit output schema documentation to the tool definition or in a separate Schema class with JsonSerializable attributes. Document all returned fields and their types.
• Document the empty input schema explicitly in tool description: 'This tool requires no input parameters. Call it to retrieve the authenticated user's session context.'
• Consider adding a parameter to optionally filter returned claims (e.g., 'include_token_expiry: bool') to give LLMs control over response verbosity, though keep the default minimal.

Inferred effective spec: 2026-07-28+.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry