Quick Start

Claude for Financial Services พร้อมเลือกให้ 3 วิธี: web UI (Cowork), CLI (Claude Code), หรือ headless (Managed Agents API)

Prerequisites

ทั้ง 3 วิธี ต้องมี:

  • Claude API key — sk-ant-... (ได้จาก console.anthropic.com)
  • Anthropic account — เข้าถึง API documentation

ตัวเลือก:

  • MCP provider credentials — Daloopa, CapIQ, FactSet, Moody’s, ฯลฯ (ใช้เต็มความสามารถ)
  • Internal MCP server — สำหรับ Managed Agents (firm-specific GL, subledger, CRM data)

วิธี 1: Cowork Plugin (Web UI) — ง่ายที่สุด

ไม่ต้องติดตั้งอะไร — login เข้า claude.com → Settings → Plugins

ขั้นตอน 1: Add Repository 

Settings → Plugins → Add plugin
↓
Paste URL: https://github.com/anthropics/claude-for-financial-services
↓
Browse marketplace for agents and verticals

(README.md line 51–55)

ขั้นตอน 2: Select Agents/Verticals

Marketplace แสดง:

  • 10 Named Agents — Pitch Agent, Market Researcher, GL Reconciler, KYC Screener, ฯลฯ
  • 7 Verticals — financial-analysis (core), investment-banking, equity-research, private-equity, wealth-management, fund-admin, operations

เลือกตามสิ่งที่ต้อง:

  • IB bank: Pitch Agent + investment-banking vertical
  • Equity research: Market Researcher + equity-research vertical
  • PE firm: KYC Screener + private-equity vertical
  • Fund admin: GL Reconciler + fund-admin vertical
  • Wealth manager: wealth-management vertical

ขั้นตอน 3: ลองใช้

หลังติดตั้ง slash commands พร้อมใช้:

/comps —— Comparable company analysis
/dcf —— DCF valuation with WACC
/lbo —— Leveraged buyout model
/cim —— Confidential Information Memorandum
/earnings —— Post-earnings update
/ic-memo —— Investment committee memo
/portfolio —— Portfolio monitoring
/financial-plan —— Financial planning
... ดูเพิ่มเติมใน /commands

(README.md line 160–213)

ตัวอย่าง: ลองครั้งแรก

สวัสดี Claude

ฉันต้องการวิเคราะห์ comps สำหรับ Acme Corporation (ACME).
ปัจจุบัน trading at $150/share, market cap $3B.

ช่วยฉันได้ไหม?

Claude จะเลือก /comps command โดยอัตโนมัติ และ:

  1. Query MCP servers (Daloopa, CapIQ, FactSet) สำหรับ peer set
  2. ดึง trading multiples (EV/EBITDA, P/E, EV/Sales, ฯลฯ)
  3. Draft summary table + outlier flags
  4. Suggest valuation range (low/median/high)

วิธี 2: Claude Code CLI — สำหรับ Developers

ต้องมี Claude Code CLI installed (install.claudecode.com)

ขั้นตอน 1: Add Marketplace 

claude plugin marketplace add anthropics/claude-for-financial-services

ขั้นตอน 2: Install Core Plugin 

# ต้องติดตั้งก่อนเสมอ — มี shared skills + MCP configs
claude plugin install financial-analysis@claude-for-financial-services

ขั้นตอน 3: Install Agents & Verticals

ติดตั้งเฉพาะที่ต้อง:

# Named agents
claude plugin install pitch-agent@claude-for-financial-services
claude plugin install market-researcher@claude-for-financial-services
claude plugin install gl-reconciler@claude-for-financial-services
claude plugin install earnings-reviewer@claude-for-financial-services
claude plugin install kyc-screener@claude-for-financial-services

# หรือ verticals (ถ้าต้องการ skills + commands แต่ไม่เฉพาะ agent workflow)
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 fund-admin@claude-for-financial-services

(README.md line 59–74)

ขั้นตอน 4: ตรวจสอบการติดตั้ง 

claude plugin list
# pitch-agent@claude-for-financial-services ✓
# market-researcher@claude-for-financial-services ✓
# financial-analysis@claude-for-financial-services ✓
# ... ฯลฯ

ขั้นตอน 5: ลองใช้ใน Session

สร้าง new session ด้วยคำสั่ง CLI:

claude start my-pitch-work

ในการสนทนา:

สวัสดี

ฉันต้องสร้าง pitch deck สำหรับ strategic acquisition ของ Acme Corp
 + competitor เข้ามา + 3 years strategic rationale
ฉันต้อง:
- Sector overview
- Comps analysis (8–10 peers)
- 5-year DCF
- LBO scenario
- Football field valuation
- PowerPoint deck (บน bank template)

เริ่มได้ไหม?

Claude จะอ่านว่า ต้อง Pitch Agent → เริ่ม workflow:

1. Confirm target, sector, situation ✓
2. Invoke sector-overview skill → draft overview
3. Query CapIQ for peers + data
4. Invoke comps-analysis skill → peer comparison
5. Invoke lbo-model, dcf-model, 3-statement-model
6. Generate football field
7. Invoke pitch-deck skill → populate PowerPoint
8. Invoke ib-check-deck → verify consistency
9. Stop & surface Excel + deck for review

ผลลัพธ์:

  • valuation.xlsx (live formulas, sensitivity tables)
  • pitch_deck.pptx (charts linked to Excel)

วิธี 3: Managed Agents API (Headless) — Production

สำหรับ internal orchestration, automation, workflow engine ของตัวเอง

Prerequisites 

# 1. Export API key
export ANTHROPIC_API_KEY=sk-ant-...

# 2. Clone repo
git clone https://github.com/anthropics/claude-for-financial-services.git
cd claude-for-financial-services

# 3. Setup Python environment
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt  # includes anthropic SDK

ขั้นตอน 1: Configure Environment 

# Copy example env
cp .env.example .env

# Edit with your credentials
vim .env

# ตัวแปรที่จำเป็น:
# ANTHROPIC_API_KEY=sk-ant-...
# GL_MCP_URL=https://your-internal-gl.company.com/mcp      (ถ้าใช้ GL Reconciler)
# SUBLEDGER_MCP_URL=https://your-internal-subledger.com/mcp (ถ้าใช้ GL Reconciler)
# CAPIQ_API_KEY=...                                         (ถ้าใช้ CapIQ)

(managed-agent-cookbooks/gl-reconciler/agent.yaml line 36–37)

ขั้นตอน 2: Deploy Agent 

# Deploy GL Reconciler
scripts/deploy-managed-agent.sh gl-reconciler

# Output:
# Created agent ID: agent_ABC123...
# System prompt: ...
# Skills: gl-recon, break-tracing, variance-commentary
# MCPs: internal-gl, subledger
# Subagents: reader.yaml, critic.yaml, resolver.yaml
# Ready for /v1/agents/{id}/run

(README.md line 80–85)

ขั้นตอน 3: Create Orchestration Loop

Deploy script สร้าง agent ใน Anthropic API แล้วตรง run script ลูปที่ handle handoff_request events:

# orchestrate.py (reference implementation)
import anthropic

client = anthropic.Anthropic()
agent_id = "agent_ABC123..."  # from deploy output

# Example: GL reconciliation handoff chain
trial_balance = """
AR: $1,000,000
AP: $500,000
... 
(unbalanced by $50,000)
"""

# Send to GL Reconciler
response = client.beta.agents.runs.run(
    agent_id=agent_id,
    model="claude-opus-4-7",
    messages=[{
        "role": "user",
        "content": f"Please reconcile this trial balance:\n{trial_balance}"
    }]
)

# Poll for completion
while response.status in ["queued", "in_progress"]:
    response = client.beta.agents.runs.retrieve(agent_id, response.id)

# Handle output
if response.status == "completed":
    # Process artifacts (logs, reconciliation report)
    print(response.messages[-1].content)
elif response.status == "requires_action":
    # Route handoff_request to next subagent (reader, critic, resolver)
    # See scripts/orchestrate.py for full loop
    pass

(scripts/orchestrate.py — README.md line 84–85)

ขั้นตอน 4: Route to Subagents

GL Reconciler มี depth-1 leaf workers:

# managed-agent-cookbooks/gl-reconciler/subagents/reader.yaml
name: gl-reconciler-reader
model: claude-opus-4-7
system:
  file: ../../plugins/agent-plugins/gl-reconciler/agents/gl-reconciler.md
  append: |
    You are the Reader. Your job: ingest GL data, validate structure, identify potential breaks.
    Report findings to the critic.

callable_agents:
  - manifest: critic.yaml

# managed-agent-cookbooks/gl-reconciler/subagents/critic.yaml
name: gl-reconciler-critic
system:
  append: |
    You are the Critic. Your job: analyze reader's findings, hypothesize root causes, 
    prioritize investigation paths. Route to resolver.

callable_agents:
  - manifest: resolver.yaml

# managed-agent-cookbooks/gl-reconciler/subagents/resolver.yaml
name: gl-reconciler-resolver
system:
  append: |
    You are the Resolver. Your job: execute root-cause analysis, 
    draft reconciliation journal entries, route for human approval.

Orchestrator loop routes handoff_request events ตามลำดับ

Verifying Installation

Cowork: ตรวจสอบ Commands 

Prompt: /comps Apple
# → Should invoke comps-analysis skill
# → Should query Daloopa/CapIQ MCP
# → Should return peer multiples table

Claude Code CLI: ตรวจสอบ Plugins 

claude plugin list | grep financial
# pitch-agent@claude-for-financial-services ✓ active
# market-researcher@claude-for-financial-services ✓ active
# financial-analysis@claude-for-financial-services ✓ active

Managed Agents: ตรวจสอบ Deploy 

curl -H "Authorization: Bearer $ANTHROPIC_API_KEY" \
     https://api.anthropic.com/v1/agents/agent_ABC123 | jq .status

# Should show: "status": "ready"

First Run Examples

Example 1: Comparative Company Analysis 

ฉันต้องการ comps analysis สำหรับ SaaS company.
- Target: Datadog (DDOG)
- Market cap: $80B
- TTM revenue: $1.8B
- TTM EBITDA margin: 20%

ช่วยฉันหา 8-10 peers พร้อม multiples ได้ไหม?

ผลลัพธ์:

  • Peer set (CloudFlare, CrowdStrike, Snyk, HashiCorp, …)
  • Trading multiples (EV/Revenue, EV/EBITDA, P/E, Price/Sales)
  • Outlier flags (CrowdStrike trading at 15x vs peer median 8x — why?)
  • Valuation range for Datadog

Example 2: DCF Valuation 

DCF model สำหรับ Acme Corp.
- Historical 3-year revenue CAGR: 25%
- Projected growth: 20%, 15%, 12%, 10%, 8% (next 5 years)
- EBITDA margin: 18% (expanding to 20% by year 3)
- Tax rate: 21%
- CapEx: 3% of revenue
- NWC: 5% of revenue
- Terminal growth: 3%
- WACC: 9% (Cost of Equity 11%, Cost of Debt 5%, D/V 30%)

สร้างโมเดล + sensitivity tables ได้ไหม?

ผลลัพธ์:

  • Excel workbook
    • Sheet 1: Inputs (assumptions)
    • Sheet 2: Historical financials + projections
    • Sheet 3: WACC calculation
    • Sheet 4: FCF bridge → Terminal Value
    • Sheet 5: DCF valuation summary
    • Sheet 6: 5×5 sensitivity table (WACC vs Terminal g)
    • Sheet 7: Football field (comps vs precedents vs DCF vs LBO)
  • All formulas live (not hardcoded)
  • Sensitivity table center cell = base case price

Common Customizations

1. Connect Your Own Data 

# Edit .mcp.json to point to internal systems
vim plugins/vertical-plugins/financial-analysis/.mcp.json

# Replace MCP URLs with your own:
{
  "mcpServers": {
    "internal-gl": {
      "type": "http",
      "url": "https://gl.company.com/mcp"    # ← your GL
    },
    "internal-crm": {
      "type": "http",
      "url": "https://crm.company.com/mcp"   # ← your CRM
    },
    ...
  }
}

# Redeploy agents
scripts/deploy-managed-agent.sh gl-reconciler

(README.md line 154–155)

2. Add Firm Context to Skills 

# Edit skill directly
vim plugins/vertical-plugins/financial-analysis/skills/dcf-model/SKILL.md

# Add firm guidelines:
# ...
# Your firm's WACC guidelines: Cost of Equity 10–12% (by sector), Cost of Debt 3–6%
# Use 10-year Treasury for risk-free rate
# Beta sourced from FactSet (adjusted for leverage)
# ...

ทั้ง Cowork + Managed Agents จะใช้ version ที่ update นี้

3. Create Custom Agent 

# Copy existing agent
cp -r plugins/agent-plugins/pitch-agent plugins/agent-plugins/my-custom-agent

# Edit system prompt
vim plugins/agent-plugins/my-custom-agent/agents/my-custom-agent.md

# Sync skills
python3 scripts/sync-agent-skills.py

# Test + validate
python3 scripts/check.py

ดูเพิ่มเติม

  • Architecture — 3-layer: verticals → agents → cookbooks
  • Principles — เหตุผลที่อยู่เบื้องหลัง design
  • Full Skill Reference — ทั้ง commands สำหรับแต่ละ vertical