MCPMarshal - Financial AI Agent

An intelligent financial AI agent that analyzes your investment portfolio, fetches real-time market data, and delivers personalized insights through multiple channels, all powered by Arcade.

Overview

MCPMarshal demonstrates how to build a production-ready AI agent using a three-layer architecture that separates concerns between data sourcing, persistent memory, and multi-channel presentation. The application showcases Arcade’s seamless OAuth handling for Gmail, Notion, and Google Docs integration.

What It Does

• Tracks Your Portfolio: Manage multiple investment portfolios with holdings, purchase prices, and quantities
• Fetches Real-Time Data: Custom MCP server retrieves live stock prices, company fundamentals, and market news
• Agentic Report Generation: LLM autonomously researches your holdings, deciding what data to fetch and where to focus attention
• Delivers Anywhere: Publish reports to Gmail, Notion, or Google Docs with one-click OAuth
• Scheduled Automation: Weekly cron jobs automatically generate and email portfolio reports
• Agentic Chat: Ask questions and watch the AI autonomously call tools to fetch the data it needs

Architecture

Layer Breakdown

Arcade Integration

Arcade is the backbone that enables seamless multi-channel publishing without the complexity of managing OAuth flows manually.

How Arcade Powers This Application

1. Tool Execution via Arcade SDK

import Arcade from "@arcadeai/arcadejs";

const arcadeClient = new Arcade({ apiKey: process.env.ARCADE_API_KEY });

// Execute any tool with automatic auth handling
const response = await arcadeClient.tools.execute({
  tool_name: "Gmail.SendEmail",
  input: {
    subject: "Weekly Portfolio Report",
    body: reportHtml,
    recipient: userEmail,
    content_type: "html",
  },
  user_id: userId,
});

2. OAuth Authorization Flow

// Check if user has authorized Gmail
const authResponse = await arcadeClient.tools.authorize({
  tool_name: "Gmail.SendEmail",
  user_id: userId,
});

if (authResponse.status === "pending") {
  // Redirect user to Arcade's OAuth consent screen
  return { authorizationUrl: authResponse.authorization_url };
}

// User is authorized, proceed with tool execution

3. Supported Integrations

Why Arcade?

• Zero OAuth Complexity: No need to register apps with Google/Notion, manage client secrets, or handle token refresh
• Unified API: Same pattern for Gmail, Notion, Google Docs, and 100+ other integrations
• User-Scoped Auth: Each user authorizes once, Arcade handles the rest
• Production Ready: Built for scale with automatic token management

Custom MCP Server

The application includes a custom MCP (Model Context Protocol) server built with arcade-mcp that provides financial data tools.

Available Tools

MCP Server Structure

mcp-server/
├── src/
│   └── financial_mcp/
│       ├── server.py              # Main MCP app entrypoint
│       ├── tools/
│       │   ├── stock_data.py      # Price and historical data
│       │   ├── company_info.py    # Company fundamentals
│       │   ├── market_news.py     # News aggregation
│       │   └── portfolio_analysis.py  # Multi-stock analysis
│       └── utils/
│           └── yfinance_client.py # Yahoo Finance wrapper
└── pyproject.toml

Deploying to Arcade

cd mcp-server
uv tool install arcade-mcp  # Install CLI if needed
arcade deploy -e src/financial_mcp/server.py

Once deployed, the MCP server is accessible through Arcade’s tool execution API, enabling the AI chat and report generation features.

Features

Portfolio Management

• Create and manage multiple portfolios
• Add holdings with ticker, quantity, and purchase price
• Real-time valuation with live market data
• Performance tracking with gain/loss calculations

Agentic Report Generation

• Autonomous research: The LLM decides what data to fetch based on portfolio composition
• Intelligent focus: Spends more time analyzing holdings with significant gains/losses
• Multi-step reasoning: Up to 15 tool calls per report for thorough analysis
• Streaming output: Watch the agent research and write in real-time
• Multi-channel publishing: Dashboard, Gmail, Notion, or Google Docs

Interactive AI Chat

• Agentic tool calling: LLM autonomously fetches market data as needed
• Natural language queries: Ask anything about your portfolio
• Multi-step reasoning: Chains tool calls to answer complex questions
• Persistent memory: Chat history saved across sessions

Scheduled Reports

• Automated weekly portfolio reports via Vercel Cron
• Emails delivered directly to your inbox
• Configurable output channels per portfolio

Tech Stack

Project Structure

├── app/
│   ├── api/
│   │   ├── ai-chat/          # AI chat streaming endpoint
│   │   └── cron/             # Vercel cron job handlers
│   ├── auth/                 # Authentication pages
│   └── dashboard/
│       ├── ai-chat/          # Interactive chat interface
│       ├── portfolio/        # Portfolio management
│       └── reports/          # Report generation & viewing
├── components/
│   ├── ui/                   # shadcn/ui components
│   └── ...                   # App components
├── lib/
│   ├── arcade.ts             # Arcade client configuration
│   ├── auth.ts               # Better Auth setup
│   └── prisma.ts             # Database client
├── mcp-server/               # Custom MCP server (Python)
└── prisma/
    └── schema.prisma         # Database schema

Getting Started

Prerequisites

• Node.js 18+
• Python 3.10+
• PostgreSQL database
• Arcade API key (Get one here)

Installation

1. Clone the repository

   git clone 
   cd mcp-project
   

2. Install Node dependencies

   npm install
   

3. Set up environment variables

   cp .env.example .env
   

   Configure the following variables:

   # Database
   DATABASE_URL="postgresql://..."
   
   # Authentication
   BETTER_AUTH_SECRET="your-secret-key"
   BETTER_AUTH_URL="http://localhost:3000"
   
   # Arcade
   ARCADE_API_KEY="your-arcade-api-key"
   
   # AI
   ANTHROPIC_API_KEY="your-anthropic-key"
   

4. Initialize the database

   npx prisma db push
   

5. Deploy the MCP server to Arcade

   cd mcp-server
   uv sync
   uv tool install arcade-mcp
   arcade deploy -e src/financial_mcp/server.py
   

6. Start the development server

   npm run dev
   

7. Open the application

   Navigate to http://localhost:3000

Environment Variables

Deployment

Vercel (Recommended)

1. Push your code to GitHub
2. Import the project in Vercel
3. Configure environment variables
4. Deploy

The application includes a vercel.json with cron job configuration for weekly reports:

{
  "crons": [
    {
      "path": "/api/cron/weekly-reports",
      "schedule": "0 9 * * 1"
    }
  ]
}

MCP Server

Deploy the MCP server to Arcade:

cd mcp-server
arcade deploy -e src/financial_mcp/server.py

Demo Walkthrough

1. Create an Account: Sign up with email/password
2. Add a Portfolio: Create a portfolio and add your holdings
3. Generate a Report: Click “Generate Report” to create an AI analysis
4. Publish to Gmail: Click “Publish” and authorize Gmail through Arcade
5. Chat with AI: Ask questions about your portfolio in the AI Chat
6. Automatic Reports: Enable scheduled reports to receive weekly emails

License

MIT

---

Built with Arcade | Next.js | Vercel AI SDK