One Open Source Project a Day (No. 59): Claude for Financial Services - Anthropic's Official AI Agent Suite for Finance

Introduction

"From watching analysts work to managing the analytical pipeline itself."

This is the 59th article in the "One Open Source Project a Day" series. Today, we are exploring Claude for Financial Services.

In the last two articles, we covered OpenAI's Symphony (an AI agent orchestration spec) and Addy Osmani's Agent Skills (an engineering discipline skill set). Today's project is different: Anthropic itself stepped into a vertical and built an AI agent suite purpose-designed for the financial industry—a comprehensive solution handling DCF valuations, LBO models, equity research reports, KYC reviews, GP/LP reporting, and general ledger reconciliation.

This is not a toy project. With 12k Stars and 1.6k Forks, it reflects workflows actively used by Wall Street banks, hedge funds, and private equity firms. More importantly, it is fully open source—you can directly see how Anthropic defines what a "financial AI agent" should look like.

What You Will Learn

• How Anthropic defines the capability boundaries for 10 specialized financial AI agents
• How 7 vertical plugins cover the full business spectrum from IB (Investment Banking) to WM (Wealth Management)
• How 11 MCP data connectors integrate professional sources like FactSet, Bloomberg (LSEG), and S&P Global
• How to deploy and use this system in Claude Code, Cowork, and Microsoft 365
• Why "outputs staged for human sign-off, never auto-executed" is the central design principle for financial AI agents

Prerequisites

• Basic familiarity with financial concepts (a passing acquaintance with DCF, LBO, PE, WM is enough—key terms are explained)
• Some experience with Claude Code or other AI coding tools
• No professional finance background required

Project Background

Project Introduction

anthropics/financial-services is an official Anthropic reference library containing Reference Agents, Skills, and Data Connectors for the financial services industry.

Its design philosophy rests on one core principle: AI drafts, humans sign off. Whether generating a Pitch Deck, automatically reconciling a general ledger, or auditing LP statements, all outputs are "staged for human sign-off"—no trades are executed, no entries are posted, no client notifications are sent automatically. This design is what enables deployment in a heavily regulated industry.

Author/Team Introduction

• Publisher: Anthropic (the company behind Claude)
• Positioning: Official reference implementation for financial institutions, demonstrating how to deploy AI agents responsibly in regulated industries
• Partners: LSEG (formerly Refinitiv) and S&P Global—two global financial data giants providing official partner plugins

Project Data

• ⭐ GitHub Stars: 12,000+
• 🍴 Forks: 1,600+
• 📝 Commits: 50
• 🤝 Official Partners: LSEG, S&P Global
• 🔌 MCP Data Connectors: 11
• 📄 License: Apache-2.0
• 🌐 Repository: anthropics/financial-services

Main Features

Core Utility

The project delivers three layers of capability:

1. 10 Named Agents: Independently deployable end-to-end workflows—e.g., the "Pitch Agent" generates a branded Pitch Deck complete with comps, precedent transactions, and LBO summary from a single command
2. 7 Vertical Plugins: Skills and commands organized by business line (IB / ER / PE / WM / Fund Admin / Operations)
3. 11 MCP Data Connectors: Integrations with Daloopa, FactSet, Morningstar, PitchBook, and other industry data sources

Use Cases

1. Investment Banking

   • Use /cim to generate a Confidential Information Memorandum, /buyer-list to build a buyer universe, and /merger-model to analyze M&A transactions.

2. Equity Research

   • During earnings season, use /earnings to instantly generate a complete research note covering the latest results, model updates, and rating maintenance; use /initiate for new coverage initiations.

3. Private Equity

   • Use /ic-memo to generate Investment Committee memos, /dd-checklist to build diligence checklists, and /portfolio to monitor portfolio companies.

4. Wealth Management

   • Use /client-review to prepare client meeting materials, /rebalance to generate portfolio rebalancing recommendations, and /tlh to perform tax loss harvesting analysis.

5. Fund Administration

   • Use the GL Reconciler agent to locate general ledger breaks and trace root causes; use the Month-End Closer agent to complete accruals, roll-forwards, and variance commentary.

Quick Start

Method 1: Claude Cowork (Simplest)

Settings → Plugins → Add Plugin
Paste: https://github.com/anthropics/claude-for-financial-services

Method 2: Claude Code CLI

# Install the core plugin first (required dependency)
claude plugin install financial-analysis@claude-for-financial-services

# Install agents for your use case
claude plugin install pitch-agent@claude-for-financial-services
claude plugin install gl-reconciler@claude-for-financial-services

# Install vertical business line plugins
claude plugin install investment-banking@claude-for-financial-services
claude plugin install equity-research@claude-for-financial-services
claude plugin install private-equity@claude-for-financial-services
claude plugin install wealth-management@claude-for-financial-services

Method 3: Managed Agents (Headless Automation)

export ANTHROPIC_API_KEY=sk-ant-...

# Deploy the general ledger reconciliation agent
scripts/deploy-managed-agent.sh gl-reconciler

# Deploy the KYC screening agent
scripts/deploy-managed-agent.sh kyc-screener

Method 4: Microsoft 365 Add-in

# Deploy into M365 using the provided installation scripts
cd claude-for-msft-365-install
# Follow README to configure Teams/Excel integration

Core Characteristics

1. 10 Named Agents

1. 7 Vertical Plugins + 2 Partner Plugins

   • Covering IB, ER, PE, WM, Fund Admin, and Operations
   • LSEG plugin: bond relative value, swap curves, FX carry, options volatility
   • S&P Global plugin: company tear sheets, earnings previews, funding digests

2. 11 MCP Data Connectors
   

   Daloopa      ←  AI-powered financial modeling data
   FactSet      ←  Comprehensive financial data and analytics
   Morningstar  ←  Fund and equity research data
   S&P Global   ←  Credit ratings and market intelligence
   LSEG         ←  Real-time and historical market data (formerly Bloomberg)
   Moody's      ←  Credit analysis and ratings
   PitchBook    ←  Private equity and venture capital data
   Chronograph  ←  PE portfolio monitoring
   Aiera        ←  AI-powered earnings call analysis
   MT Newswires ←  Real-time financial news
   Egnyte       ←  Enterprise content management (file storage)

1. Four Deployment Modes
   • Cowork (Interactive): Chat-style interface, similar to ChatGPT
   • Claude Code CLI: Developer command-line mode
   • Managed Agents (Headless): API-driven batch automation
   • Microsoft 365 Add-in: Use directly inside Teams, Excel, and Outlook

Project Advantages

Detailed Analysis

1. Project Structure: Three-Layer Architecture

financial-services/
├── plugins/
│   ├── agent-plugins/          # 10 standalone agents (each with a full workflow)
│   │   ├── pitch-agent/
│   │   ├── gl-reconciler/
│   │   ├── kyc-screener/
│   │   └── ...
│   ├── vertical-plugins/       # 7 vertical business line skill sets
│   │   ├── financial-analysis/ # Core plugin (must be installed first)
│   │   ├── investment-banking/
│   │   ├── equity-research/
│   │   ├── private-equity/
│   │   ├── wealth-management/
│   │   ├── fund-admin/
│   │   └── operations/
│   └── partner-built/          # Partner-contributed plugins
│       ├── lseg/
│       └── sp-global/
├── managed-agent-cookbooks/    # Headless deployment templates per agent
├── claude-for-msft-365-install/# Microsoft 365 provisioning tooling
└── scripts/
    ├── deploy-managed-agent.sh # One-command deployment script
    ├── check.py                # Code quality checks
    ├── validate.py             # Configuration validation
    ├── orchestrate.py          # Reference event loop (orchestrator)
    └── sync-agent-skills.py    # Skill synchronization utility

2. Command Reference: 50+ Professional Operations

financial-analysis (Core Foundation):

/comps              # Comparable company analysis
/dcf                # Discounted Cash Flow valuation
/lbo                # Leveraged Buyout model
/3-statement-model  # Three-statement financial model (P&L / B/S / CF linked)
/debug-model        # Excel model audit (formula errors, circular references)
/competitive-analysis  # Market competitive positioning
/ppt-template       # PowerPoint template generation

investment-banking:

/one-pager    # Company profile (one-page summary)
/cim          # Confidential Information Memorandum (M&A sell-side)
/teaser       # Anonymous project teaser (pre-NDA)
/buyer-list   # Potential buyer / investor universe
/merger-model # M&A analysis (accretion/dilution testing)
/process-letter  # Bid process instruction letter
/deal-tracker # Deal pipeline tracking

equity-research:

/earnings         # Post-earnings research note (model updates + rating)
/earnings-preview # Pre-earnings forecast analysis
/initiate         # New coverage initiation report
/model-update     # Periodic model update
/morning-note     # Daily morning note
/sector           # Sector thematic report
/thesis           # Investment thesis tracking and updates
/catalysts        # Catalyst calendar management
/screen           # Stock screening

private-equity:

/source          # Deal origination sourcing
/screen-deal     # Initial deal screening
/dd-checklist    # Due diligence checklist generation
/dd-prep         # DD management meeting preparation
/unit-economics  # Unit economics analysis (LTV/CAC, etc.)
/returns         # Investment return analysis (IRR / MOIC)
/ic-memo         # Investment Committee memo
/portfolio       # Portfolio company monitoring
/value-creation  # Value creation plan (100-day plan, etc.)
/ai-readiness    # Portfolio company AI readiness assessment

3. The Compliance-First "Human Sign-Off" Design

This is the single most important design decision in the entire project—and the reason it can be deployed in a regulated financial environment.

The project explicitly states that all agents:

• ✅ Will do: Draft analyst work product (reports, models, memos) for review by qualified professionals
• ❌ Will not do:
  • Make investment recommendations
  • Execute transactions
  • Bind risk positions
  • Post to ledgers
  • Approve client onboarding

This "AI drafts, human signs" design precisely fits the regulatory frameworks governing financial services. It lets AI absorb the high-volume, time-intensive analytical work while preserving the irreplaceable role of human professional judgment at the final decision point.

4. orchestrate.py: How Anthropic Writes a Production Agent Event Loop

scripts/orchestrate.py is one of the most instructive files in the entire project—it is Anthropic's reference event loop implementation, revealing the skeleton of a production-grade financial AI agent:

# Conceptual structure of orchestrate.py

while True:
    # 1. Fetch next task from queue (e.g., linear board, email trigger)
    task = fetch_next_task()

    # 2. Prepare isolated workspace and context for the task
    context = prepare_workspace(task)

    # 3. Load the relevant vertical plugin skills
    skills = load_vertical_skills(task.vertical)

    # 4. Call Claude API to perform the analysis
    result = claude.run(task, context, skills)

    # 5. Stage output for review — NEVER auto-execute
    stage_for_review(result, approver=task.assigned_analyst)

    # 6. Log trajectory (token usage, latency, model version)
    log_trajectory(task, result)

Project Links & Resources

Official Resources

• 🌟 GitHub: https://github.com/anthropics/financial-services
• 📚 Anthropic Documentation: docs.anthropic.com
• 🤝 Partner — LSEG: lseg.com
• 🤝 Partner — S&P Global: spglobal.com

Target Audience

• Financial Institution Tech Teams: Looking to rapidly build compliant AI-powered analytical workflows
• IB / PE / ER Analysts: Seeking AI tools that handle repetitive modeling and report-writing tasks
• AI Application Developers: Researching how to deploy AI agents responsibly in regulated industries
• FinTech Founders: Looking to learn from Anthropic's official design patterns for financial AI

Summary

Key Takeaways

1. Official Anthropic release: the authoritative reference implementation for financial industry AI agents
2. 10 agents × 7 verticals × 11 data sources: covering the complete financial workflow from IB to fund administration
3. "AI drafts, humans sign off": the compliance design that makes deployment in a strictly regulated industry possible
4. Four deployment modes (Cowork / Claude Code / Managed Agents / M365) for diverse technical environments
5. Fully open source under Apache-2.0: freely forkable for enterprise-internal customization

One-Line Review

This is not a fantasy about "AI entering finance"—it is Anthropic's engineered roadmap for making it happen, open-sourced for everyone to study and build on.

Find more useful knowledge and interesting products on my Homepage