HealthClawGuardrails
Active·★ 31·MIT·Updated 2026-10-22
HealthClaw Guardrails is an open-source, vendor-neutral security layer designed to sit between any AI agent and any FHIR server, providing critical safeguards for clinical data interaction. It enforces PHI redaction, maintains an immutable audit trail, implements step-up authorization, and enables human-in-the-loop validation for sensitive operations.
#Healthcare AI#FHIR#Data Security#PHI Redaction#Audit Trail#Access Control#LLM Guardrails#Compliance
01
Features
01PHI Redaction: Automatically truncates names, masks identifiers, strips addresses, and truncates birth dates to year for privacy.
02Immutable Audit Trail: Logs every read and write operation with tenant, agent, and timestamp for compliance and accountability.
03Step-up Authorization: Requires HMAC-SHA256 tokens for write operations, enhancing data integrity.
04Human-in-the-loop: Blocks clinical writes until a human confirms the action, preventing unintended consequences.
05Tenant Isolation: Scopes every query to the specific tenant and blocks cross-tenant data access.
02
Compatibility
Python
Runtime
Verified via docs
Docker
Containerization
Verified via docs
FHIR
Standard
Verified via docs
Claude
AI Platform
Verified via docs
OpenAI/Gemini
AI Platform
Verified via docs
03
Quick start
1
$ uv sync
2
$ STEP_UP_SECRET=your-secret python main.py
04
Use cases
↳Securing AI agent interactions with FHIR data by enforcing privacy and compliance guardrails.
↳Deploying a proxy in front of existing FHIR servers (e.g., Medplum, HAPI FHIR) to add a security layer without modifying the server.
↳Facilitating patient-owned data quality checks and fixes (Curatr) with provenance tracking.
↳Generating encrypted SMART Health Links for secure, patient-controlled sharing of health records.
↳Integrating with AI platforms like Claude, OpenAI, and Gemini as a plugin or adapter for healthcare applications.
05
Alternatives
platform★ 15
The approval and accountability layer for agentic AI. Identity → Policy → Approval → Trace. Try: npx sidclaw-mcp-guard demo
helio★ 14
Governance proxy for MCP: block or hold AI agent tool calls on their arguments, route approvals with the arguments shown, enforce rules the model cannot argue with, require prerequisites, cap cumulative spend across tools and servers, and audit every call. No changes to agent code or MCP servers.
scopegate★ 18
AI Access Proxy Layer — granular permission gateway for AI agents. Connect services, define scopes, get an MCP endpoint.
centian★ 15
Give your AI agents structure, guardrails, and full observability — the Agent control plane built on MCP.
Comments
Log in to leave a comment
No comments yet. Be the first!