Sage Clinical Agent 🌿

A clinical automation dashboard that simulates an autonomous healthcare agent for ambient scribing, prior authorization, patient intake, clinical decision support (CDS), and revenue cycle management (RCM) — backed by a demonstrated multi-agent system, MCP server inspector, security controls, and an Agents CLI (the Agent Studio / ADK console).

Built as a single-page React application with live mock data streams, designed for high readability in both light and dark modes.

Problem (Neurology Example for Demo)

Outpatient neurology clinics face overlapping burdens that compete for physician attention during patient visits:

Challenge	Impact
Documentation load	Physicians spend significant time on SOAP notes instead of direct patient care
Prior authorization friction	Neuroimaging (MRI/MRA), EEG, EMG/NCS, and specialty therapies require insurer portal navigation and clinical justification
Intake & eligibility	Staff manually verify coverage, copays, and appointment readiness
CDS monitoring	EEG, ICP, and seizure-risk signals need context during encounters — not as global dashboard noise
RCM leakage	Denied claims and missing documentation delay reimbursement

Traditional agent dashboards expose raw telemetry (JSON state, message buses, global alert walls) that increase cognitive load for practicing clinicians. Sage addresses this by separating the physician workspace from engineering/debug surfaces.

Solution

Sage Clinical Agent uses a master-detail split layout:

Area	Purpose
Left — Today's Queue	Persistent day schedule; select any patient; Verify only for Pending/Ineligible
Center — Clinical Workspace	Encounter workflow for the active patient only: Intake → Scribe → Prior Auth → Billing
Header — Tech Console (gear icon)	ADK multi-agent graph, MCP inspector, security posture, Agents CLI — hidden during routine care

Clinical workflow (center panel)

Intake — Eligibility and copay status for the selected patient
Scribe — Ambient transcript (collapsed by default) + AI-synthesized SOAP note as the visual focus; patient-scoped ICP/Alpha monitor
Prior Auth — Claims pipeline filtered to the active patient
Billing — RCM ledger filtered to the active patient

Tech Console (gear icon → “Back to clinic” to return)

Multi-Agent — Orchestrator + Scribe, Portal, and RCM agents; message bus and task queues
MCP Inspector — EHR, Payer, Billing, and Imaging (PACS) MCP servers with tool-call traces
Security — RBAC scope, PHI redaction log, credential vault, audit trail
Agents CLI — Named skills (/agents list, /skill run intake.verify PT-03, etc.)

Design principles

Sage & earthy neutral palette (#f7faf8 linen background, #647b6a accents)
WCAG-friendly contrast; leaf/sprout status iconography
Neurology CDS (EEG/ICP) scoped to the active patient, not pinned globally
Clinical alerts collapsed into a patient-scoped badge + drawer

Architecture

High-level layout

Sage uses a full-height master-detail shell (h-screen, fixed queue + scrollable workspace). Physicians stay in Clinical mode by default; Tech Console is one click away and does not share the queue column.

Clinical mode (default)

╔══════════════════════════════════════════════════════════════════════════╗
║  🌿 Sage Clinical Agent        Neurology · Dr. Evelyn Young      [⚙]     ║
╠════════════════════╦═════════════════════════════════════════════════════╣
║  TODAY'S QUEUE     ║  ACTIVE ENCOUNTER                                   ║
║  (320px sidebar)   ║  ─────────────────────────────────────────────────  ║
║                    ║  Arthur Pendelton · Chronic Migraine F/U · 09:00    ║
║  ┌────────────────┐║  ┌────────┐ ┌────────┐ ┌────────────┐ ┌────────┐    ║
║  │● Arthur        │║  │ Intake │ │ Scribe │ │ Prior Auth │ │Billing │    ║
║  │  ✓ Eligible    │║  └────────┘ └────────┘ └────────────┘ └────────┘    ║
║  └────────────────┘║                                                     ║
║  ┌────────────────┐║  ┌─────────────────────────────────────────────┐    ║
║  │  Sarah         │║  │         Clinical SOAP Note (center focus)    │   ║
║  │  ✓ Eligible    │║  │   S · O · A · P  —  AI draft, physician edit │   ║
║  └────────────────┘║  └─────────────────────────────────────────────┘    ║
║  ┌────────────────┐║  ▸ Ambient transcript (collapsed by default)        ║
║  │  Michael       │║  ▸ Patient monitor — ICP / Alpha (this pt only)     ║
║  │  *Pending*     │║                                                     ║
║  │  [Verify]      │║  [ Clinical Alerts ]  ← badge only if pt has alerts ║
║  └────────────────┘║                                                     ║
║  ┌────────────────┐║                                                     ║
║  │  Diana …       │║                                                     ║
║  │  Robert …      │║                                                     ║
║  └────────────────┘║                                                     ║
╚════════════════════╩═════════════════════════════════════════════════════╝
         ▲                              ▲
         │                              │
   activePatientId              EncounterWorkflow step
   (click to switch)           drives center panel content

Zone	Component	Behavior
Header	`Layout.jsx`	Branding, physician context, gear opens Tech Console
Left	`PatientQueueSidebar.jsx`	Full day list; selection highlight; Verify only for Pending/Ineligible
Right	`ClinicalWorkspace.jsx`	Per-patient Intake / Scribe / Prior Auth / Billing; SOAP is the visual anchor

Tech Console mode (gear icon → “Back to clinic”)

╔══════════════════════════════════════════════════════════════════════════╗
║  ← Back to clinic                              Tech Console (ADK debug)  ║
╠══════════════════════════════════════════════════════════════════════════╣
║  ┌─────────────┬─────────────┬─────────────┬─────────────┐               ║
║  │ Multi-Agent │ MCP Inspect │  Security   │ Agents CLI  │               ║
║  └─────────────┴─────────────┴─────────────┴─────────────┘               ║
║                                                                          ║
║   Orchestrator graph · message bus · MCP tool traces                     ║
║   RBAC · PHI log · vault · audit trail · /agents list · skills           ║
║                                                                          ║
║   (No patient queue — isolated from clinical UX)                         ║
╚══════════════════════════════════════════════════════════════════════════╝

Routing: App.jsx renders ClinicalWorkspace or AgentStudioView based on activeTab in DashboardContext — the clinical layout is never modified when debugging agents.

Application diagram

flowchart TB
  subgraph entry [Entry]
    HTML[index.html]
    MAIN[main.jsx]
  end

  subgraph state [Global State — DashboardContext.jsx]
    CTX[Mock data + simulators]
    CTX --> Queue[patientQueue + activePatientId]
    CTX --> Encounters[Per-patient SOAP / transcript / step]
    CTX --> Agents[AGENT_REGISTRY + MCP_SERVERS]
    CTX --> Sec[SECURITY_CONTROLS + PHI log]
    CTX --> CLI[runAgentCommand + AGENT_SKILLS]
  end

  subgraph shell [Shell — Layout.jsx]
    APP[App.jsx]
    LAYOUT[Header + master-detail flex]
    APP --> LAYOUT
    LAYOUT --> QSidebar[PatientQueueSidebar]
    LAYOUT --> Main[ClinicalWorkspace OR AgentStudioView]
  end

  subgraph clinical [Clinical Workspace]
    CW[ClinicalWorkspace.jsx]
    CW --> EW[EncounterWorkflow]
    CW --> PM[PatientMonitorPanel]
    CW --> SOAP[SOAP editor + ambient transcript]
    CW --> PA[PriorAuthView clinicalMode]
    CW --> RCM[RcmCdsView clinicalMode]
    CW --> AL[ClinicalAlertsDrawer]
  end

  subgraph studio [Tech Console — AgentStudioView]
    AS[Multi-Agent / MCP / Security / CLI tabs]
  end

  HTML --> MAIN
  MAIN --> CTX
  MAIN --> APP
  APP -->|activeTab clinical| CW
  APP -->|activeTab agent-studio| AS
  CTX -.->|useDashboard| shell & clinical & studio

Agent orchestration

Physicians interact with the Clinical Workspace; the Orchestrator routes work to specialist agents. Each agent calls only the MCP servers it needs — not every server.

flowchart TB
  subgraph ui [Clinical UI — physician-facing]
    Q[Today's Queue]
    CW[Clinical Workspace]
    Q --> CW
    CW --> Intake[Intake]
    CW --> Scribe[Scribe / SOAP]
    CW --> Auth[Prior Auth]
    CW --> Bill[Billing]
  end

  subgraph agents [ADK Multi-Agent — simulated]
    O[Orchestrator]
    SC[Scribe Agent]
    PO[Portal Agent]
    RC[RCM Agent]
  end

  subgraph mcp [MCP Servers — simulated]
    EHR[EHR MCP]
    PAY[Payer MCP]
    BIL[Billing MCP]
    IMG[Imaging / PACS MCP]
  end

  CW --> O
  O --> SC
  O --> PO
  O --> RC

  SC --> EHR
  PO --> PAY
  PO --> IMG
  RC --> BIL

Agent	Role	MCP servers
Orchestrator	Routes tasks to sub-agents; monitors agent registry	Registry / audit tools (no direct EHR or payer calls)
Scribe Agent	Ambient dialogue → structured SOAP note	EHR MCP — `ehr.getPatientContext`, `ehr.writeSoapDraft`
Portal Agent	Eligibility, prior auth, payer portal navigation	Payer MCP — eligibility and auth submission; Imaging MCP — `imaging.attachClinicalPacket` for auth bundles
RCM Agent	Claims status, denials, appeals	Billing MCP — `claims.fetchStatus`, `claims.createAppeal`, `billing.auditTrail`

MCP server registry

Server	Endpoint	Key tools
EHR MCP	`mcp://ehr.internal:8001`	`ehr.getPatientContext`, `ehr.writeSoapDraft`, `ehr.attachNeuroExam`
Payer MCP	`mcp://payer.clearinghouse:8002`	`payer.checkEligibility`, `payer.submitPriorAuth`, `payer.fetchDecision`
Billing MCP	`mcp://billing.internal:8003`	`claims.fetchStatus`, `claims.createAppeal`, `billing.auditTrail`
Imaging MCP	`mcp://pacs.internal:8004`	`imaging.attachClinicalPacket`, `imaging.fetchStudy`, `imaging.orderMRI`

Tech Console routing

Engineering surfaces are isolated from the clinical layout — opened via the header gear icon:

flowchart LR
  CW[Clinical Workspace] -->|gear icon| TC[Tech Console]
  TC --> MA[Multi-Agent View]
  TC --> MCP[MCP Inspector]
  TC --> SEC[Security]
  TC --> CLI[Agents CLI]

Data flow

sequenceDiagram
  participant Physician
  participant ClinicalWorkspace
  participant DashboardContext
  participant Simulators

  Physician->>ClinicalWorkspace: Select patient in queue
  ClinicalWorkspace->>DashboardContext: setActivePatient(id)
  DashboardContext->>ClinicalWorkspace: Per-patient SOAP, step, alerts

  Physician->>ClinicalWorkspace: Start ambient scribe
  ClinicalWorkspace->>DashboardContext: resetScribeSession()
  DashboardContext->>Simulators: Transcript interval, agent logs
  Simulators->>ClinicalWorkspace: Updated transcript + SOAP

  Note over DashboardContext,Simulators: MCP calls, message bus, CDS alerts<br/>run in background simulators — visible in Tech Console

Tech stack

Layer	Technology
UI	React 19
Build	Vite 8
Styling	Tailwind CSS v4 (`@theme` in `src/index.css`)
Icons	Lucide React
State	React Context (`DashboardContext.jsx`)
Lint	Oxlint
Optional API	FastAPI (`backend/`) — ADK stub, PHI/injection guards

Project structure

agy-clinical-dashboard/
├── public/
│   └── favicon.svg
├── src/
│   ├── main.jsx
│   ├── App.jsx                         # clinical vs agent-studio router
│   ├── index.css                       # Sage design tokens
│   ├── context/
│   │   └── DashboardContext.jsx        # Mock data, simulators, agent registries
│   ├── api/
│   │   └── agentApi.js                 # Optional backend client (Tech Console only)
│   └── components/
│       ├── Layout.jsx                  # Master-detail shell
│       ├── PatientQueueSidebar.jsx     # Today's Queue (left column)
│       ├── ClinicalWorkspace.jsx       # Active encounter (center)
│       ├── EncounterWorkflow.jsx       # Intake → Scribe → Prior Auth → Billing
│       ├── PatientMonitorPanel.jsx     # Patient-scoped ICP / EEG
│       ├── ClinicalAlertsDrawer.jsx    # Patient-scoped alerts badge
│       ├── PriorAuthView.jsx           # Prior auth (clinical + full modes)
│       ├── RcmCdsView.jsx              # Billing ledger (clinical + full modes)
│       └── AgentStudioView.jsx         # Tech Console shell
│           ├── MultiAgentView.jsx
│           ├── McpInspectorView.jsx
│           ├── SecurityView.jsx
│           └── AgentsCliView.jsx
├── scripts/                            # Optional runtime validation (see docs/)
├── docs/
│   └── AGENT_RUNTIME_TESTING.md
├── backend/                            # Optional FastAPI + ADK scaffold
│   ├── main.py
│   ├── agents/orchestrator/
│   ├── security/
│   └── README.md
├── implementation_plan_neurology.md
└── tasks.md

Course concepts demonstrated

This project applies Google Agent course key concepts as follows:

Key Concept	Where	How in Sage
Agent / Multi-agent (ADK)	Code	`AGENT_REGISTRY`, `MultiAgentView.jsx`, `backend/agents/orchestrator/`
MCP Server	Code	`MCP_SERVERS`, `McpInspectorView.jsx`, simulated tool-call trace
Security features	Code	`SecurityView.jsx`, PHI/injection guards in `backend/security/`
Agent skills (Agents CLI)	Code	`AgentsCliView.jsx`, `AGENT_SKILLS`, `runAgentCommand()`
Antigravity	Video	Ambient scribe + background Portal/RCM automation without physician prompting
Deployability	Video	`npm run build` → Vercel; optional `backend/Dockerfile` → Cloud Run

Code review path: npm run dev → Clinical Workspace for the physician demo → gear icon (top-right) for Tech Console.

Setup

Prerequisites

Node.js 18+ (20+ recommended)
npm 9+

No API keys are required for the frontend demo.

Install and run (frontend — primary)

git clone <your-repo-url>
cd agy-clinical-dashboard

npm install
npm run dev

Open http://localhost:5173 (or the URL printed in the terminal).

Other commands

npm run build    # Production build → dist/
npm run preview  # Preview production build locally
npm run lint     # Run Oxlint

Optional backend scaffold

The backend is not required for the clinical demo. To explore server-side PHI guards, injection detection, and CLI mirroring:

cd backend
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
uvicorn main:app --reload --port 8080

To connect only the Tech Console Agents CLI to the local API, copy .env.example to .env.local in the project root:

VITE_AGENT_API_URL=http://localhost:8080

Restart npm run dev. The clinical workspace is unchanged; only the CLI in Tech Console calls the backend.

See backend/README.md and docs/AGENT_RUNTIME_TESTING.md for API details and validation scripts.

Deploy (frontend)

npm run build

Deploy the dist/ folder to Vercel (or any static host). Typical workflow: connect the GitHub repo and auto-deploy on push to main.

Usage guide

Clinical Workspace (default view)

Select a patient in Today's Queue (left sidebar) — defaults to Arthur Pendelton.
Use the workflow stepper: Intake → Scribe → Prior Auth → Billing.
On Scribe: review the AI SOAP draft; expand Ambient transcript to start a simulated scribe session.
Verify eligibility appears only for patients with Pending or Ineligible status.
Clinical Alerts badge appears only when the active patient has high-priority alerts.

Tech Console (engineering / course demo)

Click the gear icon (top-right).
Explore Multi-Agent, MCP Inspector, Security, and Agents CLI tabs.
Try CLI commands: /agents list, /mcp ping, /security audit, /skill run intake.verify PT-03.
Click Back to clinic to return to the physician view.

Clinical persona

Element	Value
Application	Sage Clinical Agent
Physician	Dr. Evelyn Young, Neurology Specialist
Default patient	Arthur Pendelton — chronic migraine follow-up
CDS (patient-scoped)	ICP ~11 mmHg, alpha rhythms stable, EEG stream nominal
Sample CPT codes	70551 (Brain MRI), 95816 (EEG), 95886 (EMG/NCS), 64615 (Botox migraine), 99214 (office visit)

Implementation documentation

implementation_plan_neurology.md — Cardiology → neurology transition
docs/AGENT_RUNTIME_TESTING.md — Runtime validation payloads (optional backend)
tasks.md — Project task ledger

Navigation Menu

Repository files navigation

Sage Clinical Agent 🌿

Problem (Neurology Example for Demo)

Solution

Clinical workflow (center panel)

Tech Console (gear icon → “Back to clinic” to return)

Design principles

Architecture

High-level layout

Clinical mode (default)

Tech Console mode (gear icon → “Back to clinic”)

Application diagram

Agent orchestration

MCP server registry

Tech Console routing

Data flow

Tech stack

Project structure

Course concepts demonstrated

Setup

Prerequisites

Install and run (frontend — primary)

Other commands

Optional backend scaffold

Deploy (frontend)

Usage guide

Clinical Workspace (default view)

Tech Console (engineering / course demo)

Clinical persona

Implementation documentation

License

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
backend		backend
public		public
src		src
.env.example		.env.example
.gitignore		.gitignore
.oxlintrc.json		.oxlintrc.json
README.md		README.md
implementation_plan_neurology.md		implementation_plan_neurology.md
implementation_plan_v3.md		implementation_plan_v3.md
index.html		index.html
package-lock.json		package-lock.json
package.json		package.json
tasks.md		tasks.md
vite.config.js		vite.config.js

Latest commit

History

Folders and files

Repository files navigation

Sage Clinical Agent 🌿

Problem (Neurology Example for Demo)

Solution

Clinical workflow (center panel)

Tech Console (gear icon → “Back to clinic” to return)

Design principles

Architecture

High-level layout

Clinical mode (default)

Tech Console mode (gear icon → “Back to clinic”)

Application diagram

Agent orchestration

MCP server registry

Tech Console routing

Data flow

Tech stack

Project structure

Course concepts demonstrated

Setup

Prerequisites

Install and run (frontend — primary)

Other commands

Optional backend scaffold

Deploy (frontend)

Usage guide

Clinical Workspace (default view)

Tech Console (engineering / course demo)

Clinical persona

Implementation documentation

License

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages