Y.A.N.I. MCP Server

Overview

The Y.A.N.I. MCP Server exposes the full YaniPay platform to any AI model that supports the Model Context Protocol (MCP). It provides real-time access to user wallets, transactions, loyalty programs, KYC status, crypto portfolios, referral data, and a fintech knowledge base covering French regulation, DOM fiscal specifics, and competitive analysis.

What is MCP?

The Model Context Protocol is an open standard created by Anthropic that lets AI models securely interact with external data sources and tools. Instead of building custom integrations for each AI provider, MCP provides a single protocol that works with Claude, GPT, Gemini, and any compatible LLM.

For fintech, this is transformative: an AI assistant can now directly query a user's wallet balance, analyze their transaction history, check their KYC status, and provide personalized financial advice — all through a standardized, secure interface.

// 1. LLM receives user question
User: "What's my account balance?"

// 2. LLM calls MCP tool
MCP Tool: yani_get_balance({ email: "user@example.com" })

// 3. Server queries YaniPay database via Prisma
const wallets = await db.wallet.findMany({ where: { userId } })

// 4. Formatted response returned to LLM
Result: "## Soldes de Johan\n- EUROS: 1,234.56 EUR\n- YANI_COIN: 500.00 YANI"

// 5. LLM presents natural response to user
AI: "You have 1,234.56 EUR in your main wallet and 500 YANI tokens."

Architecture

The Y.A.N.I. MCP Server is a standalone Node.js TypeScript project inside the YaniPay monorepo at mcp-server-yani/. It connects to the same PostgreSQL database via Prisma ORM and serves three independent transport layers.

┌─────────────────────────────────────────────────────┐
│                   AI Clients                         │
│  Claude Code  │  ChatGPT  │  Gemini  │  Custom LLM  │
└───────┬───────┴─────┬─────┴────┬─────┴──────┬───────┘
        │             │          │            │
   ┌────▼────┐  ┌─────▼────┐  ┌─▼──────┐  ┌──▼───────┐
   │  stdio  │  │ HTTP/SSE │  │  REST  │  │  Future  │
   │  :pipe  │  │  :3100   │  │  :3101 │  │ WebSocket│
   └────┬────┘  └─────┬────┘  └──┬─────┘  └──────────┘
        │             │          │
   ┌────▼─────────────▼──────────▼────────────────────┐
   │           Y.A.N.I. MCP Server v1.2.0             │
   │                                                   │
   │  ┌─────────────┐  ┌──────────────────────────┐   │
   │  │  5 Advisory  │  │     9 Platform Tools     │   │
   │  │   Tools      │  │  (Prisma DB queries)     │   │
   │  └─────────────┘  └──────────────────────────┘   │
   │  ┌─────────────┐  ┌──────────────────────────┐   │
   │  │ 3 Resources │  │      3 Prompts           │   │
   │  └─────────────┘  └──────────────────────────┘   │
   │  ┌────────────────────────────────────────────┐   │
   │  │  Knowledge Base (4 skills, 50+ references) │   │
   │  └────────────────────────────────────────────┘   │
   └──────────────────────┬───────────────────────────┘
                          │
                ┌─────────▼─────────┐
                │   PostgreSQL DB   │
                │   (Prisma ORM)    │
                └───────────────────┘

Y.A.N.I. v3 Internal Integration

Starting with Y.A.N.I. v3.0, the agent hierarchy uses an internal MCPClientService that gives the primary YANI agent awareness of all configured MCP tools at inference time. This is separate from the external MCP server: it runs inside the agent process and injects tool names into every system prompt.

// Y.A.N.I. v3: enriched handleUserMessage pipeline
const ragContext  = ragKnowledge.getContextForPrompt(message, 3);
const mcpTools    = mcpClient.getToolNamesForPrompt();
const prompt      = systemPromptEngine.generatePrompt(
  userProfile, conversationContext, ragContext, mcpTools
);

// Metrics returned with every response
return {
  success: true,
  data: response,
  metrics: {
    ragChunksUsed:       ragContext.length,
    mcpToolsAvailable:   mcpTools.length,
    promptTokenEstimate: prompt.tokenEstimate,
    sessionMessages:     session.messageHistory.length,
  },
};

14 Tools

The server exposes 14 tools organized in two categories: advisory tools (sync, knowledge-based) and platform tools (async, database-backed).

Transports

Quick Start

The fastest way to start using the Y.A.N.I. MCP Server is with Claude Code (stdio transport):

1. Configure Claude Code— Add the MCP server to your .claude/settings.json:
   .claude/settings.jsonjson

   {
     "mcpServers": {
       "yani-advisor": {
         "command": "npx",
         "args": ["tsx", "mcp-server-yani/src/index.ts"],
         "cwd": "/path/to/yanipay-animation",
         "env": {
           "DATABASE_URL": "postgresql://user:pass@localhost:5432/yanipay"
         }
       }
     }
   }

2. Build the server:

   cd mcp-server-yani
   pnpm install
   pnpm build

3. Test with the CLI:

   # List all 14 tools
   ./yani-query.sh --list
   
   # Query financial advice
   ./yani-query.sh yani_financial_advice --query "Comment optimiser ma fiscalite a La Reunion?"
   
   # Get wallet balances
   ./yani-query.sh yani_get_balance --email user@example.com
   
   # Search knowledge base
   ./yani-query.sh yani_search_knowledge --query "fiscalite DOM"

Use Cases

The Y.A.N.I. MCP Server enables innovative AI-powered fintech experiences:

Next Steps