SentryMCP

Multi-domain compliance agent — ask audit questions in plain English or by voice, route to the right subsystem via MCP, get a structured verdict from AWS Bedrock, hear it aloud with Amazon Polly.

| | | |---|---| | Stack | Python · FastAPI · MCP · Bedrock · Polly | | Domains | FinTech · Healthcare · HR | | Pattern | Clean / Hexagonal architecture (ports & adapters) |

🎬 Demo: (Add GIF here — screen record the voice-to-verdict flow)

Table of Contents

• Why
• What
• Architecture
• Prerequisites
• Quick Start with Docker
• Setup & Installation
  • 1. Clone the repo
  • 2. Create virtual environment
  • 3. Install dependencies
  • 4. Configure environment variables
  • 5. Run the server
• Running Modes
  • Local mode (no AWS)
  • Bedrock mode (AWS)
• Usage
• Sample Output
• Project Layout
• API Reference
• Tests
• Troubleshooting
• Documentation

Why

Compliance teams jump between siloed systems — ledgers, patient records, HR policies. Checks are slow, inconsistent, and hard to audit. SentryMCP gives one interface for natural-language compliance triage with explainable JSON verdicts (CLEARED · FLAGGED · ACTION_REQUIRED).

What

• Text or voice audits through a built-in web UI (GET /)
• MCP routing to domain-specific tools (mock data in this POC)
• LLM reasoning with Bedrock + free-model fallback chain
• Multi-turn chat — follow-ups like "What about their flagged transactions?" keep context
• Voice-to-voice — browser speech recognition in, Polly TTS out

30-second architecture view:

Operator → Web UI / API → FastAPI → AnalyzeVoiceUseCase
    → MCP server (FinTech | Healthcare | HR)
    → Bedrock LLM → structured verdict → (optional) Polly TTS

Architecture

Core business logic in the center; HTTP, MCP, and AWS at the edges. New industries = new MCP server, no core rewrite needed.

HLD — System Overview

LLD — Sequence Flows

| Flow | Diagram | |------|---------| | Text chat | | | Voice + Polly | | | Conversation memory | |

Prerequisites

Make sure you have the following installed before starting:

| Requirement | Version | Check | |---|---|---| | Python | 3.11+ | python --version | | pip | latest | pip --version | | Git | any | git --version | | AWS CLI (optional, Bedrock mode only) | v2+ | aws --version |

No AWS account? Run in Local mode — no cloud setup needed.

Quick Start with Docker

The fastest way to run SentryMCP — no Python setup needed.

Prerequisites: Docker Desktop installed and running.

Local mode (no AWS)

# 1. Clone
git clone https://github.com/krsatyam99/SentryMCP.git
cd SentryMCP

# 2. Copy env file (defaults to local mode — no AWS needed)
cp .env.example .env        # macOS/Linux
copy .env.example .env      # Windows

# 3. Build and run
docker compose up --build

Open http://localhost:8000/ — done. ✅

Bedrock + Polly mode (AWS)

# Edit .env first
LLM_PROVIDER=bedrock
POLLY_ENABLED=true
AWS_ACCESS_KEY_ID=your_key
AWS_SECRET_ACCESS_KEY=your_secret

# Then run
docker compose up --build

Dev mode (hot reload)

docker compose --profile dev up sentrymcp-dev

Server runs on http://localhost:8001/ and reloads on code changes.

Useful Docker commands

# Run in background
docker compose up -d

# View logs
docker compose logs -f

# Stop
docker compose down

# Rebuild after dependency changes
docker compose up --build

Setup & Installation

1. Clone the repo

git clone https://github.com/krsatyam99/SentryMCP.git
cd SentryMCP

2. Create virtual environment

Windows (PowerShell):

python -m venv .venv
.venv\Scripts\activate

macOS / Linux:

python -m venv .venv
source .venv/bin/activate

You should see (.venv) in your terminal prompt.

3. Install dependencies

pip install -r requirements.txt
pip install -e .

4. Configure environment variables

Copy the example env file and fill in your values:

Windows:

copy .env.example .env

macOS / Linux:

cp .env.example .env

Then open .env and set your values:

# ── LLM Provider ─────────────────────────────────────────
# Use "local" for free OpenRouter fallback (no AWS needed)
# Use "bedrock" for AWS Bedrock (requires AWS credentials below)
LLM_PROVIDER=local

# ── Voice Output ─────────────────────────────────────────
# Set to true to enable Amazon Polly TTS (requires AWS credentials)
POLLY_ENABLED=false

# ── AWS Credentials (only needed for bedrock or Polly) ───
AWS_ACCESS_KEY_ID=your_access_key_here
AWS_SECRET_ACCESS_KEY=your_secret_key_here
AWS_DEFAULT_REGION=us-east-1

# ── Bedrock Model ─────────────────────────────────────────
BEDROCK_MODEL_ID=amazon.nova-micro-v1:0

5. Run the server

# Windows
$env:PYTHONPATH = "src"
python -m uvicorn agentai.adapters.inbound.api.app:app --host 0.0.0.0 --port 8000

# macOS / Linux
PYTHONPATH=src python -m uvicorn agentai.adapters.inbound.api.app:app --host 0.0.0.0 --port 8000

You should see:

INFO:     Started server process [xxxx]
INFO:     Application startup complete.
INFO:     Uvicorn running on http://0.0.0.0:8000

Open http://localhost:8000/ in Chrome or Edge.

Running Modes

Local mode (no AWS)

No AWS account needed. Uses a free OpenRouter model fallback chain.

LLM_PROVIDER=local
POLLY_ENABLED=false

Note: Free models may be rate-limited or return slower responses.

Bedrock mode (AWS)

Full production-like setup with AWS Bedrock for LLM reasoning and Polly for voice output.

Step 1 — Configure AWS credentials:

# Windows PowerShell
$env:AWS_ACCESS_KEY_ID="your_access_key"
$env:AWS_SECRET_ACCESS_KEY="your_secret_key"
$env:AWS_DEFAULT_REGION="us-east-1"

Or use the AWS CLI:

aws configure

Step 2 — Set env vars:

LLM_PROVIDER=bedrock
POLLY_ENABLED=true
BEDROCK_MODEL_ID=amazon.nova-micro-v1:0

Step 3 — Ensure your IAM user has these permissions:

• bedrock:InvokeModel
• polly:SynthesizeSpeech

Usage

| URL | Purpose | |-----|---------| | http://localhost:8000/ | Web UI (chat + voice) | | http://localhost:8000/docs | Interactive API explorer |

Try these sample queries:

# FinTech
Check account ACC-991A for possible fraud
What about their flagged transactions?

# Healthcare
Check patient PAT-204B for compliance risk

# HR
What is the data privacy policy?
Summarize the whistleblower policy

Sample Output

{
  "verdict": "FLAGGED",
  "confidence": 0.91,
  "domain": "fintech",
  "account": "ACC-991A",
  "reason": "Offshore transaction TX-302 ($95,000) flagged for manual review. Compliance status: UNDER_REVIEW.",
  "compliance_note": "CRITICAL: Escalate to compliance officer within 24 hours.",
  "mcp_source": "fintech_server"
}

Verdict values:

| Verdict | Meaning | |---|---| | CLEARED | No compliance issues detected | | FLAGGED | Suspicious activity — review recommended | | ACTION_REQUIRED | Immediate escalation needed |

Project Layout

SentryMCP/
├── backend/
│   └── mcp_servers/          # Domain MCP plugins (stdio transport)
│       ├── fintech_server.py
│       ├── healthcare_server.py
│       └── hr_server.py
├── src/agentai/
│   ├── core/                 # Entities, ports, use cases (pure Python)
│   ├── adapters/
│   │   ├── inbound/          # FastAPI routes
│   │   └── outbound/         # MCP client, Bedrock, Polly adapters
│   └── static/               # Web UI (index.html)
├── docs/                     # Technical guide (DOCX) + diagrams
├── tests/
├── Dockerfile
├── docker-compose.yml
├── .dockerignore
├── .env.example
├── requirements.txt
└── setup.py

API Reference

Full interactive docs at http://localhost:8000/docs

POST /analyze

{
  "industry": "fintech",
  "query": "Check account ACC-991A for possible fraud"
}

Response: structured verdict JSON (see Sample Output)

Tests

pytest

# With coverage
pytest --cov=src/agentai tests/

Troubleshooting

claude or uvicorn not recognized:

# Refresh PATH in current session
$env:PATH += ";C:\Users\<you>\.local\bin"

Port 8000 already in use:

netstat -ano | findstr :8000
taskkill /PID <PID> /F

AWS credentials error:

• Confirm AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY are set correctly
• Check IAM permissions include bedrock:InvokeModel and polly:SynthesizeSpeech
• Switch to LLM_PROVIDER=local to test without AWS

Free model rate limit (429):

• Expected behavior — the system falls back automatically through the model chain
• Switch to LLM_PROVIDER=bedrock for reliable responses

MCP server not found:

• Ensure you ran pip install -e . from the repo root
• Check PYTHONPATH=src is set before running uvicorn

Documentation

For deep-dive technical reference (architecture decisions, MCP plug-in guide, full API, AWS setup, demo script):

📄 docs/SentryMCP-Technical-Guide.docx

Open in Microsoft Word or Google Docs.

Package: cross-industry-voice-dataguard v0.1.0
Built with: Python · FastAPI · MCP · AWS Bedrock · Amazon Polly