An HTTP API and MCP server for fast PHI de-identification with full reversibility. Strip all 18 HIPAA identifiers, route to any consumer LLM, and re-identify on return. Self-hosted or cloud.
Consumption-based credits • Self-hosted or managed • No minimums
De-identify clinical text in milliseconds, send it to any model, and get back a response with all patient context restored.
Drop-in replacement for OpenAI and Anthropic endpoints. Same request format, same streaming — MedScrub handles de-identification transparently.
Model Context Protocol interface for agent workflows. Use MedScrub as a tool in any AI pipeline — de-identify, process, re-identify in one call.
PHI is stripped before data leaves your infrastructure. Use GPT-4o, Claude, Gemini, or local models — no BAA required with AI providers.
Send clinical text with PHI. Get back an AI response with PHI restored. MedScrub handles de-identification, LLM routing, and re-identification transparently.
curl -X POST https://api.medscrub.ai/v1/chat \
-H "Authorization: Bearer $API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-4o",
"messages": [{
"role": "user",
"content": "Summarize this patient:
John Smith, DOB 3/15/1979,
MRN 847291, HbA1c 7.8%,
BP 138/88, BMI 31.2"
}]
}'"[PATIENT_001], [AGE] yo,
HbA1c 7.8%, BP 138/88, BMI 31.2"MedScrub sits between your application and any AI provider. PHI never reaches the model.
Your app sends PHI → MedScrub strips it → Clean data reaches the LLM → Response re-identified on return
Monitor and manage your self-hosted MedScrub platform — PHI Proxy, NER Engine, CDR Server, Session Store, and Database all running on your infrastructure.


Strips all 18 HIPAA identifiers with high accuracy on structured FHIR data. Deterministic, auditable, 100% reversible.
Backend systems integration for Epic and athenahealth. Read patient data, write clinical notes back via FHIR R4.
Self-hosted clinical data repository. Your patient data in a FHIR-native store you control — not locked in a vendor's cloud.
Route to GPT-4o, Claude, Gemini, or any OpenAI-compatible endpoint. Switch models without changing your code.
Model Context Protocol support for agent workflows. Use MedScrub as a tool in your AI pipeline.
Deploy on your infrastructure. Docker, Kubernetes, or bare metal. Air-gapped deployments supported for maximum security.
Any AI feature that touches clinical data — made HIPAA-compliant in one line
Differential diagnosis, treatment suggestions, drug interactions — powered by any LLM
Generate notes, summaries, discharge instructions, or patient communications from de-identified data
Chatbots for intake, symptom checking, health coaching, or care navigation — safely personalized
AI-assisted ICD-10, CPT, or SNOMED coding from clinical documentation
AI-powered patient routing, referral management, or care plan generation
Use de-identified production data in staging environments without HIPAA risk
Go from idea to HIPAA-compliant MVP in weeks, not months. No enterprise contracts or BAA negotiations.
Integrate clinical AI into existing products without rebuilding your compliance infrastructure.
Self-host for maximum control. Run AI workloads on clinical data without PHI leaving your network.
PHI de-identified before AI processing
On structured FHIR data, deterministic approach
Full re-identification for EHR write-back
Air-gapped deployment, your infrastructure