Enterprise Financial Compliance Audit Framework
A production-grade MCP server that automates financial compliance audits by combining LLMs with structured analysis, enabling policy validation, risk profiling, and automated reporting.
README
Enterprise Financial Compliance Audit Framework
AI-Native Compliance Infrastructure — A production-grade MCP server that combines Large Language Models with structured financial analysis to automate enterprise compliance workflows.
Table of Contents
- Overview
- Architecture
- Project Structure
- Features
- MCP Tools Reference
- Quick Start
- Deployment
- API Usage
- Security Considerations
- Technical Highlights
Overview
This project implements a modular compliance audit system built on the Model Context Protocol (MCP). It enables AI agents to perform sophisticated financial compliance analysis through a standardized tool interface.
Key Capabilities
- Policy Validation Engine: Multi-layer transaction analysis against configurable business rules
- Risk Profiling: Vendor risk assessment with external data enrichment
- Automated Reporting: Executive-grade PDF reports with statistical anomaly detection
- LLM-Driven Orchestration: Groq LLM integration for intelligent tool selection
- Real-time Dashboard: Interactive web visualization with role-based access control
Use Case
Enterprise finance teams need to audit thousands of transactions across multiple vendors, checking for:
- Blacklisted vendor exposure
- Category spending limit violations
- Pending transaction aggregation risks
- Historical vendor risk patterns
This system automates that analysis pipeline end-to-end.
Architecture
┌─────────────────────────────────────────────────────────────────────────────┐
│ ORCHESTRATION LAYER │
│ ┌─────────────────────────────────────────────────────────────────────┐ │
│ │ Host Client (host.py) │ │
│ │ ┌──────────────┐ ┌──────────────┐ ┌──────────────────────────┐ │ │
│ │ │ Rate Limiter│ │ RBAC Engine │ │ Audit Logger │ │ │
│ │ │ (1 req/sec) │ │(Role-based) │ │ (RotatingFileHandler) │ │ │
│ │ └──────────────┘ └──────────────┘ └──────────────────────────┘ │ │
│ │ ┌─────────────────────────────────────────────────────────────┐ │ │
│ │ │ Groq LLM Integration (llama-3.3-70b) │ │ │
│ │ │ • Tool Selection • Argument Generation │ │ │
│ │ └─────────────────────────────────────────────────────────────┘ │ │
│ └─────────────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────────────┘
│
│ HTTP/MCP Protocol
▼
┌─────────────────────────────────────────────────────────────────────────────┐
│ MCP SERVER LAYER │
│ ┌─────────────────────────────────────────────────────────────────────┐ │
│ │ FastMCP Server (server.py) │ │
│ │ Port: 8000 │ │
│ │ ┌─────────────────────────────────────────────────────────────┐ │ │
│ │ │ TOOLS (5) │ │ │
│ │ │ ┌─────────────────┐ ┌─────────────────┐ ┌────────────────┐ │ │ │
│ │ │ │validate_policy │ │flag_high_value │ │get_vendor_risk │ │ │ │
│ │ │ │ _tools() │ │ _transactions()│ │ _profile() │ │ │ │
│ │ │ └─────────────────┘ └─────────────────┘ └────────────────┘ │ │ │
│ │ │ ┌─────────────────┐ ┌─────────────────┐ │ │ │
│ │ │ │enrich_transaction│ │generate_audit_ │ │ │ │
│ │ │ │ _context() │ │ _report() │ │ │ │
│ │ │ └─────────────────┘ └─────────────────┘ │ │ │
│ │ └─────────────────────────────────────────────────────────────┘ │ │
│ │ ┌─────────────────────────────────────────────────────────────┐ │ │
│ │ │ RESOURCES (1) │ │ │
│ │ │ db://schema (Database Introspection) │ │ │
│ │ └─────────────────────────────────────────────────────────────┘ │ │
│ │ ┌─────────────────────────────────────────────────────────────┐ │ │
│ │ │ PROMPTS (1) │ │ │
│ │ │ compliance_audit_prompt (System Template) │ │ │
│ │ └─────────────────────────────────────────────────────────────┘ │ │
│ └─────────────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────────────┘
│
│ SQL/JSON/Filesystem
▼
┌─────────────────────────────────────────────────────────────────────────────┐
│ DATA LAYER │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ ┌─────────────────┐ │
│ │ SQLite │ │ Static JSON │ │ Log Files │ │ Generated │ │
│ │ (audit.db) │ │ (risk_db.json)│ │(JSONL format)│ │ Reports │ │
│ │ │ │ │ │ │ │ (PDF/PNG) │ │
│ │ • transactions│ │ • vendor_risk│ │ • audit_trace│ │ │ │
│ │ • audit_hist │ │ _scores │ │ │ │ │ │
│ │ • vendor_risk│ │ │ │ │ │ │ │
│ └──────────────┘ └──────────────┘ └──────────────┘ └─────────────────┘ │
└─────────────────────────────────────────────────────────────────────────────┘
│
│ HTTP
▼
┌─────────────────────────────────────────────────────────────────────────────┐
│ VISUALIZATION LAYER │
│ ┌─────────────────────────────────────────────────────────────────────┐ │
│ │ FastAPI Dashboard (dashboard.py) │ │
│ │ Port: 8000 (shared) │ │
│ │ ┌─────────────────┐ ┌─────────────────┐ ┌─────────────────────┐ │ │
│ │ │ JWT Auth │ │ Plotly Charts │ │ KPI Metrics │ │ │
│ │ │ (Bearer Token) │ │ (Interactive) │ │ (Real-time) │ │ │
│ │ └─────────────────┘ └─────────────────┘ └─────────────────────┘ │ │
│ │ ┌──────────────────────────────────────────────────────────────┐ │ │
│ │ │ Security Middleware: Rate Limiting, HSTS, X-Frame-Options │ │ │
│ │ └──────────────────────────────────────────────────────────────┘ │ │
│ └─────────────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────────────┘
Project Structure
MCP-Server/
├── server.py # MCP server entry point (73 lines)
├── host.py # LLM orchestrator with audit loop
├── dashboard.py # FastAPI web dashboard with security
├── db/
│ ├── __init__.py
│ └── connection.py # SQLite with WAL mode, connection pooling
├── models/ # Pydantic schemas for validation
├── prompts/
│ └── compliance_audit.py # MCP prompt templates
├── resources/
│ └── schema.py # db://schema resource implementation
├── tools/ # Modular MCP tools
│ ├── __init__.py
│ ├── validation_tools.py # Policy validation, high-value flagging
│ ├── enrichment.py # Vendor risk enrichment
│ └── reporting.py # PDF report generation
├── utils/ # Utility functions
├── templates/
│ └── dashboard.html # Jinja2 dashboard template
├── requirements.txt
├── Dockerfile
└── .gitignore # Properly excludes .env, *.db, etc.
Features
1. Multi-Layer Policy Validation
# Validates against:
- Blacklisted vendors (configurable set)
- Category spending limits (Legal: $25K, IT: $40K, Travel: $20K)
- Pending transaction aggregation ($50K threshold)
2. LLM-Driven Tool Selection
The host client uses Groq's Llama 3.3 70B model to:
- Select appropriate audit tools based on natural language queries
- Generate structured arguments for tool execution
- Reason about compliance findings
3. Statistical Anomaly Detection
# Z-score based anomaly detection
if z_score > 2.0:
flag_statistical_anomaly()
perform_root_cause_analysis()
4. Risk Enrichment Pipeline
Raw Finding → Vendor Lookup → Risk Score Calculation → Enriched Output
│ │ │ │
▼ ▼ ▼ ▼
Transaction risk_db.json Weighted Algorithm Final Report
5. Executive Reporting
- PDF Generation: ReportLab-based professional reports
- Visualizations: Matplotlib charts (severity distribution, exposure trends, heatmaps)
- Markdown Export: Structured text for further processing
6. Security Features
- JWT Authentication: Role-based tokens with expiration
- Rate Limiting: 60 requests/minute default, 20/minute for sensitive endpoints
- Security Headers: HSTS, X-Frame-Options, X-Content-Type-Options, X-XSS-Protection
- HTTPS Enforcement: Automatic redirect in production mode
MCP Tools Reference
Tool: validate_transaction_policies
Validates transactions against enterprise policies.
Input: None (queries all transactions)
Output:
[
{
"severity": "HIGH",
"issue": "Blacklisted Vendor",
"vendor": "Fraudulent Corp",
"category": "IT",
"amount": 50000.00
}
]
Tool: flag_high_value_transactions
Flags transactions exceeding monetary threshold.
Input:
{
"min_amount": 10000.0
}
Output:
[
{
"vendor": "LegalEdge LLP",
"amount": 25000.00,
"category": "Legal"
}
]
Tool: get_vendor_risk_profile
Retrieves historical transaction statistics for a vendor.
Input:
{
"vendor_name": "OfficeSupply Co"
}
Output:
{
"vendor": "OfficeSupply Co",
"transaction_count": 45,
"total_spend": 125000.00,
"avg_transaction": 2777.78,
"max_transaction": 15000.00
}
Tool: enrich_transaction_context
Adds external risk context to vendor data.
Input:
{
"vendor_name": "OfficeSupply Co"
}
Output:
{
"vendor": "OfficeSupply Co",
"past_issues": true,
"conflict_of_interest": true,
"regulatory_flag": false,
"risk_score": 82
}
Tool: generate_audit_markdown_report
Generates comprehensive compliance report.
Input:
{
"findings": [...],
"flow_id": "AUDIT_a1b2c3d4",
"user_id": "auditor_01",
"role": "partner"
}
Output: Markdown string with embedded KPIs and chart references
Resource: db://schema
Returns database schema for introspection.
Output:
{
"transactions": [
{"column": "vendor", "type": "TEXT"},
{"column": "amount", "type": "REAL"},
{"column": "category", "type": "TEXT"},
{"column": "status", "type": "TEXT"}
]
}
Prompt: compliance_audit_prompt
System prompt for LLM-based compliance analysis.
Content:
You are a Senior Financial Compliance Analyst.
- Group findings by severity.
- Provide recommended actions for HIGH severity.
- Maintain professional tone.
- Use only supplied data.
Quick Start
Prerequisites
- Python 3.12+
- Groq API key
Installation
# Clone repository
git clone https://github.com/nikhil-sharma-dotcom/MCP-Server.git
cd MCP-Server
# Install dependencies
pip install -r requirements.txt
# Set environment variables
export GROQ_API_KEY="your_groq_api_key"
export SECRET_KEY="your_jwt_secret_key"
export PORT=8000
export ENVIRONMENT=development # or 'production' for HTTPS enforcement
Running Locally
# Terminal 1: Start MCP Server
python server.py
# Terminal 2: Start Dashboard
uvicorn dashboard:app --host 0.0.0.0 --port 8000
# Terminal 3: Run Audit Workflow
python host.py
Expected Output
MCP ready.
--- Enriched Audit Report ---
[LLM-generated compliance analysis]
--- Final Audit Markdown Report ---
# Enterprise Financial Compliance Report
## Executive KPIs
- **Total Risk Exposure:** $1,250,000.00
- **Total Findings:** 15
...
Deployment
Docker (Single Container)
# Build image
docker build -t mcp-audit-server .
# Run container
docker run -p 8000:8000 \
-e GROQ_API_KEY="your_key" \
-e SECRET_KEY="your_secret" \
-e PORT=8000 \
-e ENVIRONMENT=production \
mcp-audit-server
Docker Compose (Recommended)
# docker-compose.yml
version: '3.8'
services:
mcp-server:
build: .
ports:
- "8000:8000"
environment:
- GROQ_API_KEY=${GROQ_API_KEY}
- SECRET_KEY=${SECRET_KEY}
- PORT=8000
- ENVIRONMENT=production
volumes:
- ./data:/app/data
Cloud Deployment (AWS)
# ECR Push
docker build -t mcp-audit-server .
docker tag mcp-audit-server:latest $AWS_ACCOUNT_ID.dkr.ecr.$REGION.amazonaws.com/mcp-audit-server:latest
docker push $AWS_ACCOUNT_ID.dkr.ecr.$REGION.amazonaws.com/mcp-audit-server:latest
# ECS/Fargate Deployment
# Use AWS Console or Terraform for production deployment
API Usage
MCP Server HTTP Endpoint
# List available tools
curl http://localhost:8000/mcp/tools
# Call a tool
curl -X POST http://localhost:8000/mcp/tools/flag_high_value_transactions \
-H "Content-Type: application/json" \
-d '{"min_amount": 10000}'
Dashboard Endpoints
# Health check
curl http://localhost:8000/health
# Get audit history (with JWT)
curl http://localhost:8000/secure-history \
-H "Authorization: Bearer <token>"
# View dashboard
curl http://localhost:8000/ \
-H "Authorization: Bearer <token>"
Programmatic Client
from fastmcp import Client
import asyncio
async def audit_client():
async with Client("http://localhost:8000/mcp") as session:
# List tools
tools = await session.list_tools()
print(f"Available tools: {[t.name for t in tools]}")
# Call tool
result = await session.call_tool(
"flag_high_value_transactions",
{"min_amount": 5000}
)
print(result)
asyncio.run(audit_client())
Security Considerations
Implemented Security Measures
| Feature | Implementation |
|---|---|
| Authentication | JWT with role-based claims |
| Rate Limiting | slowapi (60/min default, 20/min sensitive) |
| HTTPS | Automatic redirect in production |
| Security Headers | HSTS, X-Frame-Options, X-Content-Type-Options, X-XSS-Protection |
| Secrets Management | Environment variables only |
| Logging | RotatingFileHandler with 5MB rotation |
Environment Variables
# Required
export GROQ_API_KEY="your_groq_api_key"
export SECRET_KEY="your_jwt_secret_min_32_chars"
# Optional
export PORT=8000 # Server port
export ENVIRONMENT=development # or 'production'
export DB_PATH="audit.db" # Database path
Known Limitations
- SQLite: File-based database limits horizontal scaling
- SQL Injection: Table name validation needed for schema resource
- No CSRF Protection: Stateless JWT doesn't require CSRF, but cookie-based auth would
The Refactoring Story
"I took a 665-line monolithic MCP server and refactored it into a modular architecture with:
- 73-line clean entry point
- Proper package separation (tools/, db/, models/)
- Environment-based configuration
- Comprehensive security middleware
- Professional logging with rotation"
This demonstrates technical debt management and continuous improvement — key senior-engineer traits.
Technical Highlights
1. Modular MCP Server
# server.py - Clean entry point (73 lines)
from fastmcp import FastMCP
from tools import ALL_TOOLS
from resources import database_schema_resource
from prompts import compliance_audit_prompt
mcp = FastMCP("Enterprise_Audit_Framework", version="2024-11-05")
# Register all tools with descriptions
for tool in ALL_TOOLS:
mcp.tool(
name=tool.__name__,
description=TOOL_DESCRIPTIONS.get(tool.__name__)
)(tool)
mcp.resource("db://schema")(database_schema_resource)
mcp.prompt()(compliance_audit_prompt)
2. Database Connection Management
# db/connection.py - WAL mode for better concurrency
def get_write_connection():
conn = sqlite3.connect(DB_PATH, check_same_thread=False, timeout=10)
conn.execute("PRAGMA journal_mode=WAL;") # Write-Ahead Logging
conn.execute("PRAGMA foreign_keys=ON;")
conn.row_factory = sqlite3.Row
return conn
3. Security Middleware
# dashboard.py - Production-ready security
class SecurityHeadersMiddleware(BaseHTTPMiddleware):
async def dispatch(self, request, call_next):
response = await call_next(request)
response.headers["Strict-Transport-Security"] = "max-age=31536000"
response.headers["X-Content-Type-Options"] = "nosniff"
response.headers["X-Frame-Options"] = "DENY"
return response
# Rate limiting
limiter = Limiter(key_func=get_remote_address, default_limits=["60/minute"])
4. LLM-Driven Orchestration
# host.py - Intelligent tool selection
response = groq_client.chat.completions.create(
model="llama-3.3-70b-versatile",
messages=[{"role": "user", "content": "Run full policy validation audit."}],
tools=groq_tools, # Dynamically generated from MCP tools
tool_choice="auto"
)
5. Professional Logging
# RotatingFileHandler for production
logger = logging.getLogger("audit")
handler = RotatingFileHandler(
"audit_trace.log",
maxBytes=5_000_000, # 5MB rotation
backupCount=5
)
logger.addHandler(handler)
Contributing
This is a personal portfolio project. While not actively seeking contributions, feedback and suggestions are welcome via GitHub issues.
License
MIT License - See LICENSE for details.
Acknowledgments
- FastMCP for the excellent MCP framework
- Model Context Protocol for the protocol specification
- Groq for LLM inference
Contact
Nikhil Sharma
GitHub: @nikhil-sharma-dotcom
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.