fhir-r6-guardrails

PHI redaction for FHIR data over MCP is handled automatically during read, search, and write operations. The guardrails intercept agent requests, evaluate R6 permissions, and strip protected health information before returning responses.

Two-phase write for FHIR resources separates proposal from commitment using step-up authorization. The propose phase validates permissions and tenant isolation, while the commit phase finalizes the write only after additional authorization checks pass.

Both FHIR R4 US Core v9 and FHIR R6 ballot3 are supported for reading, searching, and writing. The guardrails evaluate R6 Permission resources and enforce tenant isolation across both FHIR versions.

Immutable audit trails for FHIR access are enforced automatically by logging every agent interaction through MCP. The guardrails record read, search, and write events with PHI redaction status and permission evaluation results.

Step-up authorization for FHIR writes requires elevated permission validation during the two-phase propose and commit workflow. It prevents unauthorized agent writes by demanding additional access verification before final commitment.

R6 Permission resources are evaluated directly to enforce tenant isolation and access control for FHIR data. The guardrails interpret these resources during MCP requests to ensure compliant multi-tenant agent access.