CFO Agent Engine - Multi-Agent Workflow Demo

A Notion-driven multi-agent system for Fractional CFO operations. Shows how Claude Code orchestrates specialist agents that read tasks from Notion and autonomously execute financial workflows.

What It Does

• CEO Dispatch Agent reads incoming CFO tasks and routes them to the right specialist
• Financial Analysis Agent generates P&L reports, cash flow forecasts, variance analysis
• Reconciliation Agent matches bank transactions against accounting entries
• Triage Agent classifies uncategorized transactions using historical patterns
• Report Generation Agent formats deliverables (PDFs, slides, executive summaries)
• Review Agent QA checks all outputs before client delivery

How to Run

pip install -r requirements.txt
streamlit run app.py

Architecture

Notion Task Board
    |
    v
CEO Dispatch Agent (classifies + routes)
    |
    +---> Financial Analysis Agent
    +---> Reconciliation Agent
    +---> Triage Agent
    |
    v
Report Generation Agent (formats deliverables)
    |
    v
Review Agent (QA validation)
    |
    v
Client Delivery (Notion + attachments)

Production Architecture

In production, each agent runs as a Claude Code subprocess with:

• Dedicated system prompt and tool access
• MCP connections to Notion API, QuickBooks API, and AWS
• Dagster pipeline integration for data engineering tasks
• Execution logging and error handling
• Automatic retry and escalation

Files

• app.py - Streamlit UI with task board, pipeline execution, and agent registry
• agents.py - Multi-agent dispatch engine and execution simulation
• sample_data/tasks.json - Sample CFO tasks (P&L, reconciliation, forecasts)
• sample_data/agents.json - Agent definitions and capabilities
• workflow.md - Mermaid diagram of the pipeline flow