Agent + validated components + CLI verification pipeline.
Build medical software with an agent that verifies every formula.
A Claude Code plugin with pre-validated clinical components, a medprotocol CLI for logic verification, and 1,198 tests across 22 suites. Ship health tech that's correct by construction — not just generated.
Use it with Claude Code
One command. That's it.
Open Claude Code in any project directory, paste this command, and the agent loads the full medical protocol — 14 skills, CLI verification, privacy hooks, and automated QA.
Fetch and follow https://medical-protocol.vercel.app/medical-protocol/providers/claude-code/install.mdThe agent’s toolkit
Validated components + CLI. The agent does the rest.
Featured component
Vital Signs Monitor
6 vital parameters with validated ranges, FHIR R4 Bundle output, and LOINC codes. The agent adapts this to your project's stack — not a template you copy-paste.
The agent builds it. The CLI verifies it.
Winter's formula, anion gap, SOFA scoring, DKA resolution, ASCVD risk, CHA₂DS₂-VASc, CKD-EPI eGFR, KDIGO staging, KFRE — 1,198 unit tests across 22 suites plus the medprotocol CLI cross-check every calculation the agent generates.
medprotocol CLI
Independent command-line calculator with 9 commands: ABG, DKA, vitals, BMI, PaFi, water balance, cardiology, sepsis, and CKD. The agent uses it to cross-verify generated code.
Vital Signs
BP, HR, RR, temperature, SpO2/FiO2 with validated ranges, color-coded alerts, and complete FHIR R4 Bundle output with 8 LOINC codes.
ABG Analyzer
Primary disorder classification, compensation status, anion gap, delta ratio. Winter’s formula, Henderson-Hasselbalch, and 5 respiratory compensation formulas.
Pulse Oximetry
Real-time SpO2 and heart rate monitoring with animated waveform visualization and configurable alarm thresholds.
Breathing Rate
Respiratory rate monitoring with chest expansion charting, configurable RPM ranges, and danger/warning callbacks.
BMI Calculator
WHO classification with 4 clinical categories. Supports imperial and metric units with automatic conversion.
Water Balance
Intake/output tracking with insensible loss formula (weight × 12 mL), endogenous water generation, and net balance computation.
Claude Code Plugin
14 skills with reference docs, 4 privacy hooks, 10 context files. Install once and the agent routes clinical requests to the right workflow automatically.
Protocol Audit
Score your project against clinical compliance criteria. The agent checks theming, privacy hooks, QA pipeline, FHIR output, and test coverage.
Cardiology Suite
ASCVD risk calculator, HEART score, and CHA₂DS₂-VASc scoring with evidence-based thresholds and typed outputs.
DKA Monitor
Hourly DKA parameter tracking with automatic resolution criteria detection and configurable alert thresholds.
Sepsis Monitor
Sepsis-3 SOFA scoring with 6-organ dysfunction tracking, qSOFA screening, lactate clearance, and hour-1 bundle compliance.
PaO2/FiO2 Ratio
Computes PaO2/FiO2 ratio and returns Berlin Definition ARDS classification (mild/moderate/severe) via typed callback.
Nephrology / CKD
CKD-EPI 2021 eGFR, KDIGO staging with GFR/albuminuria risk heatmap, Tangri 4-variable KFRE for kidney failure risk, treatment eligibility (RASi, SGLT2i, finerenone), and progression monitoring with eGFR slope tracking. 563 test cases.
EHR System
Electronic health records with patient management, clinical history, structured encounters, and FHIR R4 interoperability.
What you can build
Describe the app. The agent builds it.
The agent adapts validated clinical components to your stack. You focus on product logic, not medical math.
Telehealth Platforms
Tell the agent "build a telehealth dashboard with vitals and ABG." It generates verified components adapted to your app.
EHR Integrations
FHIR R4 output with LOINC codes baked in. The agent generates interoperable components — no HL7 spec reading required.
Clinical Decision Support
Validated clinical calculators (ASCVD, HEART, CHA2DS2-VASc) with evidence-based thresholds and typed outputs.
Custom Health Apps
The agent composes any combination of clinical modules. Wellness apps, research tools, patient dashboards — all with verified logic.
How it works
Three steps. Verified clinical code.
Install the protocol. Describe your app. The agent builds it, the CLI verifies it, the tests prove it.
Step 1 — Install the protocol
Create a project directory, run claude, and paste the install command. The agent loads the plugin with 14 skills, the medprotocol CLI, 4 privacy hooks, and an automated QA pipeline.
Step 2 — Describe your app
Describe your app in technical or clinical terms. The agent routes to the right skill, installs components via the CLI, adapts them to your stack, and runs medprotocol to verify every calculation.
Step 3 — Iterate and ship
Your app runs locally with hot reload. The clinical logic is already verified by the agent. Modify layouts, wire up APIs, add features — focus on your product, not medical math.
Built for compliance
Privacy enforced at the agent level, not just promised.
Hooks block all outbound data commands automatically. Patient data stays local — enforced by the agent, not just by policy.
Privacy hooks block exfiltration
PreToolUse hooks block git push, curl POST, scp, npm publish, and docker push. Enforced at the agent level before any command runs.
Build-time vs runtime boundary
The agent generates code at build-time. Patient data exists only at runtime. Clean architectural boundary — clinical data never touches the LLM.
Inspectable shell scripts
Every privacy hook is a readable shell script, not black-box middleware. Audit them, extend them, add your own compliance rules.
- Agent hooks block all outbound data commands
- medprotocol CLI cross-verifies calculations
- FHIR R4 output with LOINC codes
- Automated QA pipeline after every generation
- 1,198 unit tests across 22 suites covering clinical logic
- TypeScript strict mode throughout
Common questions
Straight answers, no hand-waving.
"Why not just ask ChatGPT to write the clinical logic?"
You can — and you'll get code that handles common cases. But clinical logic fails at edge cases: triple acid-base disorders, DKA resolution when glucose normalizes but ketones persist. Medical Protocol's agent generates AND verifies against a CLI with 1,198 tests across 22 suites covering exactly these cases.
"What's the tech stack?"
React 19, Next.js (app router), shadcn/ui v4+, Tailwind CSS, TypeScript strict mode. Available as a Claude Code plugin with 14 skills and 4 privacy hooks. Components install via npx medical-ui-cli — same model as shadcn/ui.
"How are clinical calculations validated?"
Two layers: 1,198 Vitest unit tests across 22 suites cover every clinical formula with edge cases. Plus the medprotocol CLI (9 commands) independently verifies calculations — the agent cross-checks its own output against the CLI.
"What about patient data privacy?"
Privacy hooks block all outbound data commands at the agent level — git push, curl POST, scp, npm publish. Enforced automatically when the plugin is installed. The hooks are inspectable shell scripts.
"Can I use this commercially?"
The components and CLI are open source. You need a Claude subscription for the AI agent, but the generated code and clinical logic are yours to deploy and distribute.
"How does this compare to building from scratch?"
The agent has the clinical domain expertise your team doesn't — validated formulas, normal ranges, classification criteria, edge cases. You skip months of medical research and formula validation.
Get started
Ready in minutes, not months.
You'll need Node.js 18+ and a Claude Code subscription. Takes about 3 minutes.
Before you start
You need a Claude Code subscription from Anthropic. It's a CLI tool that generates and modifies code using AI. The medical components and clinical logic are free and open source. Claude Code
Copy this command and paste it into Claude Code
Fetch and follow https://medical-protocol.vercel.app/medical-protocol/providers/claude-code/install.md
What happens after you paste the command
Agent loads the protocol
Claude loads the plugin with 14 skills, the medprotocol CLI (9 calculators), 4 privacy hooks, and an automated QA pipeline. Your project is configured with clinical infrastructure in ~30 seconds.
Describe your application
Describe what you need in technical or clinical terms. The agent selects components, adapts them to your stack, and generates the code.
Agent verifies the output
The agent runs tsc --noEmit, the medprotocol CLI (9 commands) with known inputs, and compares results against the generated code. Clinical logic is cross-verified before you see it.
Dev server starts
Your app opens at localhost:3000 with hot reload. All components are in your project directory — verified clinical logic, ready to customize.
This tool is designed to assist healthcare professionals and has not been certified or cleared as a medical device by the FDA, EMA, or any regulatory body. It is not intended for clinical diagnosis or as a substitute for professional medical judgment.
- AI-assisted: Interface components and analysis features may be generated or powered by artificial intelligence. Always verify AI outputs independently.
- Data privacy: All patient data is processed and stored locally on your device. No clinical data is transmitted to external servers.
- Clinical responsibility: Verify all values, calculations, and recommendations before making treatment decisions.