aks129/HealthClawGuardrails — Open-source guardrails between AI agents and FHIR clinical data — PHI redaction, immutable audit, st…

Description and use

**Brief Summary**
HealthClawGuardrails (aks129/HealthClawGuardrails) is an open‑source framework that inserts “guardrails” between AI agents and FHIR‑based clinical data. It provides PHI redaction, immutable audit logging, step‑up authentication, and tenant isolation via a modular MCP server with adapters for OpenAI and Gemini, enabling safe, repeatable multi‑agent workflows in health‑tech environments.

**Value**

• **Compliance‑first data handling** – automatic PHI scrubbing and tamper‑evident audit trails keep interactions with patient data within HIPAA‑compatible boundaries.
• **Secure, context‑aware access** – step‑up authentication and per‑tenant isolation prevent cross‑tenant data leakage and enforce least‑privilege principles.
• **Reusable agent pipelines** – the MCP server and language‑agnostic adapters let developers stitch together isolated prompts, tool calls, and memory stores into deterministic, version‑controlled workflows, reducing ad‑hoc scripting and operational risk.

**Practical Adoption Path**

1. **Pilot the MCP server** – spin up the Docker‑compose deployment (or run the Python package) in a dev sandbox and point it at a test FHIR endpoint.
2. **Integrate an LLM adapter** – choose the OpenAI or Gemini adapter (or implement a custom one) and configure the API keys and authentication policies.
3. **Define guardrail policies** – use the provided YAML/JSON schema to declare PHI redaction rules, audit log destinations, and step‑up auth triggers.
4. **Wrap existing agents** – replace direct LLM calls in your code with the MCP client SDK/CLI, letting the server mediate every request.
5. **Scale to production** – deploy the server on Kubernetes or a managed VM, enable TLS, configure persistent audit storage (e.g., immutable object store), and enable tenant‑level RBAC.

**Production Readiness**

• **Recent activity** (last commit 2026‑07‑05) and a modest but growing community (22 stars, 5 forks) indicate active maintenance.
• **Strong technical signals**: Python implementation, clear API/SDK/CLI surface, and support for two major LLM providers make integration straightforward.
• **Compliance‑ready features** (PHI redaction, immutable logs, step‑up auth) address the biggest regulatory hurdles for health‑AI deployments.
• **Risks** are limited to standard OSS concerns—final license verification, deeper security review of third‑party dependencies, and confirmation of long‑term maintainers.

Overall, HealthClawGuardrails is a high‑readiness candidate for organizations that need to operationalize AI agents on sensitive clinical data while retaining auditability, isolation, and regulatory compliance.

Problem

aks129/HealthClawGuardrails helps turn isolated prompts and tools into repeatable agent workflows.

Outcome

High for an OSS candidate: recent activity, adoption, and ecosystem signals are strong enough for a serious pilot.

Workflow

• coordinate multi-agent workflows
• add tool-use pipelines
• standardize agent memory

Evidence

Open-source guardrails between AI agents and FHIR clinical data — PHI redaction, immutable audit, step-up auth, tenant isolation. MCP server + OpenAI/Gemini adapters. A healthclaw.io project.

Source