VIGO — The First MCP Server for Financial Regulatory Intelligence

Regulatory Intelligence Expert · Bilingual AI Agent for Hong Kong SFC Regulations

By OrphicOne · English & 繁體中文 · Continuously Updated Knowledge Base

What is VIGO?

VIGO (遠見 · 洞察 · 治理 · 優化) is a specialized financial regulatory intelligence AI agent — and the industry's first MCP server dedicated to financial regulatory intelligence in any jurisdiction.

At its core, VIGO is a regulatory intelligence expert: deeply versed in Hong Kong SFC regulations, enforcement history, licensing requirements, and AML/CFT rules. But VIGO is more than a compliance encyclopedia. It is a business-aware compliance navigator — an AI that understands how compliance intersects with business strategy, capital structures, and market reality.

Compliance is the floor. Business is the ceiling. VIGO helps you build both.

V.I.G.O. — The Four Dimensions

Why VIGO instead of asking Claude/ChatGPT directly?

Asking ChatGPT about SFC regulations is like asking a smart friend — they roughly know. Asking VIGO is like consulting a senior compliance partner who knows every circular, every enforcement action, every licensing nuance, and always connects compliance back to your business objectives.

Related MCP server: eu-regulations

Quick Start

Claude (Web, Desktop, Mobile)

Settings → Connectors → Add custom connector → paste:

https://lsoatzzwpltpydwyfqqv.supabase.co/functions/v1/vigo-mcp

Claude will automatically invoke VIGO when you ask SFC compliance questions.

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "vigo": {
      "type": "streamableHttp",
      "url": "https://lsoatzzwpltpydwyfqqv.supabase.co/functions/v1/vigo-mcp"
    }
  }
}

VS Code

Add to .vscode/mcp.json:

{
  "servers": {
    "vigo": {
      "type": "streamableHttp",
      "url": "https://lsoatzzwpltpydwyfqqv.supabase.co/functions/v1/vigo-mcp"
    }
  }
}

Claude Code

claude mcp add --transport http vigo https://lsoatzzwpltpydwyfqqv.supabase.co/functions/v1/vigo-mcp

ChatGPT Desktop

{
  "mcpServers": {
    "vigo": {
      "url": "https://lsoatzzwpltpydwyfqqv.supabase.co/functions/v1/vigo-mcp"
    }
  }
}

Tools

VIGO provides 6 tools (4 read-only query tools + 2 analytical tools):

query_regulation

Ask any question about SFC regulations. Covers all license types, VATP/ASPIRe framework, AML/CFT, exams, CPD/CPT, and more. Bilingual — responds in whatever language you ask.

search_enforcement

Search SFC enforcement actions — fines, suspensions, bans, and prosecutions.

get_license_guide

Comprehensive guide for any SFC license type: exams, capital, fit-and-proper, CPD.

latest_updates

Latest SFC regulatory updates: circulars, enforcement, VATP developments, consultations.

compliance_check

Perform a compliance health check for a licensed corporation. Input company profile, get a multi-dimensional compliance risk assessment.

risk_assessment

Assess compliance risk of a planned business activity. Returns Red/Yellow/Green risk rating with regulatory basis and recommended actions.

Example Queries

Licensing & Exams

What exams do I need for an SFC Type 9 (Asset Management) license?
→ Complete exam structure, pass marks, exemption conditions, official links

Enforcement Intelligence

Search for SFC enforcement actions related to AML in 2025
→ Matching cases with companies, penalties, regulatory basis, official notices

Bilingual Regulatory Updates

SFC最新的通函和監管動態是什麼？
→ 最新通函列表，含參考編號、日期、主題、官方文件鏈接（繁體中文）

Compliance Health Check

Run a compliance check for a Type 1 + Type 9 firm with 3 ROs, 25 staff, 2B HKD AUM
→ FRR capital assessment, RO configuration review, regulatory focus areas

Risk Assessment

We want to launch a tokenized fund product for retail investors in Hong Kong
→ Red/Yellow/Green rating, regulatory basis, recommended actions, timeline

Fee & Market Intelligence

What are the current RO retainer fees and shell company prices?
→ Structured fee benchmarks from vigo_fee_benchmarks with latest market data

Knowledge Base

VIGO's intelligence is powered by a structured, continuously updated knowledge base with 6,381 verified knowledge chunks covering 26 years of SFC regulatory history:

Data Sources

Official Sources (6 endpoints + HKMA)

• SFC Circulars, Enforcement News, Consultation Papers

• VATP/Virtual Asset regulatory updates

• Press Releases and policy announcements

• HKMA Joint Circulars (auto-filtered for SFC-relevant content)

Foundational Documents (35 + 5 Thematic Reports)

• 12 Codes + 16 Guidelines + 3 Laws (SFO, SFF, AMLO) + 2 Handbooks + 2 HKEX Listing Rules

• 5 SFC Thematic Inspection Reports (Cybersecurity, Sponsor Business, Prime Services, Securities Margin Financing, Alternative Liquidity Pools)

Industry Intelligence (17 sources)

• International Law Firms (6): Charltons, Deacons, KWM, Sidley Austin, Linklaters, Clifford Chance

• Local Law Firms (3): ONC Lawyers, Stevenson Wong, Paradox Management

• Big 4 (4): Deloitte HK, PwC HK, EY HK, KPMG HK

• Compliance Consultants (2): CompliancePlus, Waystone

• Compliance Tech (2): Heinbro, BBCIncorp

v7.0 Structured Data (76 records)

Coverage

• All SFC license types (Type 1–13) and VATP licensing

• AML/CFT guidelines and enforcement case history

• Exam requirements, CPD/CPT rules, fit-and-proper criteria

• FRR capital requirements and operational compliance

• HKEX Main Board and GEM Listing Rules

• Cross-border: Stock Connect, MRF Fund Mutual Recognition, WMC

Quality Assurance

Verification Grading

• 🥇 GOLD (authority ≥ 90): 4,155 records (65%) — SFC official, HKMA, HKEX

• 🥈 SILVER (authority ≥ 65): 2,226 records (35%) — Law firms, Big 4, consultants

• 🥉 BRONZE (authority < 65): 0 records

Quality Framework

• Three-layer deduplication: L1 ref_id exact match → L2 SHA-256 content hash → L3 semantic similarity > 0.92

• Six-point content validation on every chunk before upload

• Five-tier authority scoring: Primary legislation (100) → Senior speeches/Big 4 (80-90) → Specialist firms (68-75) → Brokers/media (50-55) → Social/AI (5-25)

• Bilingual: every entry exists in English and Traditional Chinese

v7.0 Intelligent Engine

Six-Layer Retrieval Pipeline

Query → [Layer 1] Query Router (9 route types)
      → [Layer 2] Hybrid Retrieval (Dense cosine@0.25 top20 + BM25 sparse top15)
      → [Layer 3] RRF Fusion (K=60, structured results get 1.5× boost)
      → [Layer 4] Score Fusion (rerank×0.50 + verification×0.20 + freshness + authority)
      → [Layer 5] Diversity Filter (min 3 sources, max 2 per source, cross-contamination penalty)
      → [Layer 6] Dynamic TopK (6-8 results based on query complexity)
      → Answer with citations

Dual-Model Routing

Built-in fallback: if GPT-4o-mini fails → auto-switch to DeepSeek.

9 Query Route Types

Each route type has optimised search weights (freshness, authority) and strategy (structured_lookup / semantic / multi_query / hybrid):

capital_requirement · license_market · enforcement_lookup · fee_benchmark · latest_update · comparison · how_to · general · unknown

Six-Layer Anti-Hallucination

1. LICENSE_TYPE_MAP hardcoded — all SFC exam subjects verified against HKSI website 2026

2. System Prompt strict instructions — never supplement specific numbers from training data

3. Reranking filter — demote non-matching license type documents

4. Verification grade weighting — GOLD > SILVER > BRONZE

5. Cross-contamination penalty — -0.05 for mismatched license documents

6. Diversity filter — force minimum 3 different sources

Iceberg Strategy (Response Philosophy)

Autonomous Operations

v7.0 achieves fully autonomous operations with zero manual intervention:

pg_cron Scheduled Tasks (4)

GitHub Actions Three-Tier Scheduling

Monthly GitHub Actions usage: ~350 minutes (well within free tier of 2,000 minutes).

Collection Pipeline

auto_collect_upload.py v7.0.3 (6,292 lines) supports 14+ modes:

Architecture

┌─────────────────────────────────────────────┐
│  MCP Clients                                │
│  Claude · ChatGPT · Cursor · VS Code       │
│  Claude Code · Any MCP Client               │
└──────────────────┬──────────────────────────┘
                   │ MCP Protocol (Streamable HTTP)
                   ▼
┌─────────────────────────────────────────────┐
│  VIGO MCP Server v2.1 (6 tools, 582 lines) │
│  Supabase Edge Function (Deno TypeScript)   │
└──────────────────┬──────────────────────────┘
                   │
    ┌──────────────┼──────────────┐
    ▼              ▼              ▼
 OpenAI        Supabase       DeepSeek + GPT-4o-mini
 Embeddings    PostgreSQL     Dual-Model Routing
 (1536-dim)    pgvector       (with auto-fallback)
               + BM25 FTS

┌─────────────────────────────────────────────┐
│  vigo-chat v5.2 (1,032 lines)               │
│  Query Router (9 types) → Hybrid Retrieval  │
│  → RRF Fusion → Score Fusion → Diversity    │
│  → Dual-Model → Engram Memory → Response    │
└─────────────────────────────────────────────┘

┌─────────────────────────────────────────────┐
│  vigo-ingest v6.1.2 (458 lines)             │
│  Auto-classify → Semantic chunk → Verify    │
│  → Grade (GOLD/SILVER/BRONZE) → Store       │
│  → Auto-generate quick questions             │
└─────────────────────────────────────────────┘

┌─────────────────────────────────────────────┐
│  Autonomous Operations                      │
│  pg_cron (4 tasks) + GitHub Actions (3)     │
│  auto_collect_upload.py v7.0.3 (6,292 lines)│
│  14+ modes · Fully autonomous · Zero manual │
└─────────────────────────────────────────────┘

Edge Functions

Database

Transport: Streamable HTTP (MCP standard)
Authentication: None required
Runtime: Supabase Edge Functions (Deno TypeScript)

Note: The MCP server is the open, stateless interface for any AI client. On the OrphicOne platform, VIGO also integrates Engram — a long-term memory system that stores conversation history, tracks user preferences, and adapts responses based on learning progress. Engram uses semantic retrieval (vector similarity) over past dialogues to deliver increasingly personalised compliance guidance over time.

Distribution

Version History

About

OrphicOne is a Hong Kong-based regulatory intelligence platform for financial services.

VIGO is OrphicOne's AI-powered regulatory intelligence agent — the industry's first and currently only MCP server dedicated to financial regulatory intelligence in any jurisdiction.

Positioning

Among 16,000+ MCP servers globally, VIGO occupies a unique position:

Global MCP Ecosystem (16,000+ servers)
├── Developer Tools        ████████████████  thousands
├── Productivity           ██████████        hundreds
├── Data & Analytics       ████████          hundreds
├── Financial Reg Intelligence █                 VIGO (only one)
└── HK SFC Reg Intelligence    █                 VIGO (industry first)

Data Moat

• 26 years of SFC regulatory history (2000–2026)

• 6,381 verified knowledge chunks with three-tier quality grading

• 76 structured records across 4 specialised tables (enforcement, regulations, fees, market)

• 35 foundational documents + 5 thematic inspection reports

• 17 industry sources crawled monthly with sub-page extraction

• HKMA joint circulars + HKEX listing rules cross-referenced

• Three-layer deduplication + six-point content validation

• v7.0 intelligent engine with dual-model routing, self-correction, and autonomous operations

• World-Class benchmarks: Recall@5 80%, MRR 0.800, duplication 1.0%, GOLD rate 64.8%

• Knowledge that took years of domain expertise to curate — not easily replicated

Philosophy

VIGO is built on the belief that compliance is not a cost — it is a competitive advantage. Great compliance doesn't slow business down; it provides the foundation for sustainable growth, client trust, and market credibility.

Compliance is the floor (The Floor), ensuring safety. Business is the ceiling (The Ceiling), achieving breakthroughs. Grounded in compliance, fluent in capital. Compliance as shield, business as spear.

Roadmap

Privacy

VIGO is a read-only regulatory intelligence service. We do not collect personal information, store conversation data, or require authentication. All query tools are marked readOnlyHint: true. See our Privacy Policy.

VIGO provides regulatory information query services only and does not constitute legal or financial advice. Users should treat VIGO's answers as reference and consult professional compliance advisors or lawyers when necessary.

Support

• Email: contact@orphicone.com

• Issues: GitHub Issues

• Website: orphicone.com

License

MIT