Zyfai DeFi MCP Server

A production-ready Model Context Protocol (MCP) server that exposes Zyfai DeFi APIs through 15 powerful tools. Built on top of the @zyfai/sdk and supports Streamable HTTP transport (MCP 2024-11-05+) with complete portfolio management, analytics, and DeFi opportunities discovery.

You can make use of the official Zyfai mcp server deployed here or run your own.

Features

• 15 MCP Tools for complete DeFi workflow
• Portfolio Management - Track positions across all chains
• Opportunities Discovery - Find conservative and aggressive yield opportunities
• Analytics & Metrics - TVL, volume, wallet analytics, and more
• Earnings Tracking - Onchain earnings, daily earnings, APY history
• User Data - Transaction history, positions, first topup info
• Multi-Chain Support - Base (8453), Arbitrum (42161), Plasma (9745)
• Streamable HTTP transport - Modern unified /mcp endpoint (MCP 2024-11-05+)
• Session-based with Mcp-Session-Id header support
• Express.js server with CORS support
• Production-ready with PM2 process management
• Comprehensive error handling & TypeScript
• Built with @modelcontextprotocol/sdk
• Powered by @zyfai/sdk

Transport Update: SSE to Streamable HTTP

This server has been updated from the deprecated SSE transport to Streamable HTTP (MCP specification 2024-11-05+):

Old (Deprecated)	New (Current)
/sse (GET) + /messages (POST)	/mcp (unified endpoint)
Separate endpoints for streaming and messages	Single endpoint handles all operations
Session managed via query params	Session managed via Mcp-Session-Id header

Benefits of Streamable HTTP

• Simpler architecture: Single unified endpoint for all MCP operations
• Better infrastructure compatibility: Works well with proxies, CDNs, and HTTP/2
• Enhanced resumability: Better error handling and session recovery
• Stateless option: Servers can operate statelessly if desired

Available Tools

Protocol (1 tool)

• get-available-protocols - Get available DeFi protocols and pools for a specific chain on Zyfai

Opportunities Discovery (2 tools)

• get-conservative-opportunities - Get safe (low risk) DeFi opportunities suitable for conservative investors
• get-aggressive-opportunities - Get degen (high-risk, high-reward) yield strategies for aggressive investors

Analytics & Metrics (6 tools)

• get-tvl - Get total value locked (TVL) across all Zyfai accounts
• get-volume - Get total volume across all Zyfai accounts
• get-active-wallets - Get active wallets for a specific chain on Zyfai
• get-smart-wallet-by-eoa - Get smart wallets associated with an EOA address
• get-rebalance-frequency - Get rebalance frequency/tier for a wallet
• get-apy-per-strategy - Get APY per strategy for a specific chain

User Data (3 tools)

• get-positions - Get all active DeFi positions and portfolio for a user's wallet address
• get-history - Get transaction history for a wallet with pagination
• get-first-topup - Get the first topup (deposit) information for a wallet

Earnings (3 tools)

• get-onchain-earnings - Get onchain earnings for a wallet including total, current, and lifetime earnings
• get-daily-earnings - Get daily earnings for a wallet within a date range
• get-daily-apy-history - Get daily APY history for a wallet

Project Structure

zyfai-mcp-server/
├── index.ts                              # Main Streamable HTTP server entry point
├── index-stdio.ts                        # STDIO server for Claude Desktop
├── proxy-server.ts                       # Proxy for remote Streamable HTTP server
├── src/
│   ├── config/
│   │   └── env.ts                        # Environment configuration
│   ├── services/
│   │   └── zyfai-api.service.ts          # Zyfai SDK wrapper service
│   ├── tools/
│   │   ├── index.ts                      # Tool registration
│   │   ├── protocol.tools.ts             # Protocol tools (1 tool)
│   │   ├── opportunities.tools.ts        # Opportunities tools (2 tools)
│   │   ├── analytics.tools.ts            # Analytics tools (6 tools)
│   │   ├── user-data.tools.ts            # User data tools (3 tools)
│   │   └── earnings.tools.ts             # Earnings tools (3 tools)
│   ├── routes/
│   │   └── http.routes.ts                # Streamable HTTP routes
│   └── middleware/
│       └── index.ts                      # Middleware (logger, error handler)
├── package.json                          # Project dependencies
├── tsconfig.json                         # TypeScript configuration
├── ecosystem.config.cjs                  # PM2 configuration
├── Dockerfile                            # Docker configuration
├── setup-domain.sh                       # Domain setup script
└── build/                                # Compiled JavaScript output

Client Integration

Using with Claude Code

If you'd like to add zyfai mcp server under your claude code, execute the below in a separate terminal, not under a claude code session:

# Using Streamable HTTP transport (recommended)
claude mcp add --transport http zyfai-agent https://mcp.zyf.ai/mcp

Using with Claude Desktop

For remote Streamable HTTP server:

{
  "mcpServers": {
    "zyfai-defi": {
      "command": "npx",
      "args": ["mcp-remote", "https://mcp.zyf.ai/mcp"]
    }
  }
}

Using with Cursor

Add to your Cursor MCP settings:

{
  "mcpServers": {
    "zyfai-defi": {
      "url": "https://mcp.zyf.ai/mcp",
      "transport": "http"
    }
  }
}

Using with Cursor / Other MCP Clients

The server uses Streamable HTTP transport, making it accessible from web browsers and HTTP clients:

import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StreamableHTTPClientTransport } from "@modelcontextprotocol/sdk/client/streamableHttp.js";

// Initialize MCP client with Streamable HTTP transport
const transport = new StreamableHTTPClientTransport(
  new URL("https://mcp.zyf.ai/mcp")
);

const client = new Client(
  { name: "my-app", version: "1.0.0" },
  { capabilities: {} }
);

await client.connect(transport);

// List available tools
const tools = await client.listTools();
console.log(
  "Available tools:",
  tools.tools.map((t) => t.name)
);

// Call a tool
const result = await client.callTool({
  name: "get-conservative-opportunities",
  arguments: { chainId: 8453 },
});

https://mcp.zyf.ai/mcp

Testing with MCP Inspector

You can test the server using the official MCP Inspector:

npx @modelcontextprotocol/inspector

Then enter the endpoint URL: https://mcp.zyf.ai/mcp

Building LLM-Powered DeFi Apps with Zyfai MCP

For developers building AI-powered DeFi agents that can autonomously discover yield opportunities, analyze portfolios, and provide intelligent recommendations, here's how to integrate the Zyfai MCP server with your LLM application.

Installation:

# For Anthropic Claude
npm install @modelcontextprotocol/sdk @anthropic-ai/sdk

# For OpenAI GPT
npm install @modelcontextprotocol/sdk openai

# Or use pnpm
pnpm add @modelcontextprotocol/sdk openai

Complete Example - AI DeFi Yield Optimizer:

This example shows how to build an LLM agent that uses Zyfai MCP server to create an intelligent DeFi assistant. Choose between OpenAI or Anthropic based on your preference.

import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StreamableHTTPClientTransport } from "@modelcontextprotocol/sdk/client/streamableHttp.js";
import OpenAI from "openai";

/**
 * Initialize Zyfai MCP Client with Streamable HTTP
 */
async function initializeZyfaiMCP() {
  const transport = new StreamableHTTPClientTransport(
    new URL("https://mcp.zyf.ai/mcp")
  );

  const client = new Client(
    {
      name: "defi-ai-agent",
      version: "1.0.0",
    },
    {
      capabilities: {},
    }
  );

  await client.connect(transport);
  console.log("Connected to Zyfai MCP Server (Streamable HTTP)");

  return client;
}

/**
 * DeFi AI Agent with OpenAI - Combines LLM reasoning with Zyfai MCP tools
 */
class DeFiAIAgent {
  private zyfaiClient: Client;
  private openai: OpenAI;
  private availableTools: any[];

  constructor(zyfaiClient: Client, openaiApiKey: string) {
    this.zyfaiClient = zyfaiClient;
    this.openai = new OpenAI({ apiKey: openaiApiKey });
    this.availableTools = [];
  }

  /**
   * Initialize agent by discovering available Zyfai tools
   */
  async initialize() {
    const toolsResponse = await this.zyfaiClient.listTools();

    // Convert MCP tool schema to OpenAI function calling format
    this.availableTools = toolsResponse.tools.map((tool) => ({
      type: "function",
      function: {
        name: tool.name,
        description: tool.description,
        parameters: tool.inputSchema,
      },
    }));

    console.log(
      `Agent initialized with ${this.availableTools.length} Zyfai tools`
    );
  }

  /**
   * Execute a tool call via Zyfai MCP
   */
  async executeTool(toolName: string, toolInput: any) {
    console.log(`Executing: ${toolName}`, toolInput);

    const result = await this.zyfaiClient.callTool({
      name: toolName,
      arguments: toolInput,
    });

    // Extract text content from MCP response
    const content = result.content
      .filter((item: any) => item.type === "text")
      .map((item: any) => item.text)
      .join("\n");

    return content;
  }

  /**
   * Chat with the AI agent - it can use Zyfai tools automatically
   */
  async chat(userMessage: string, conversationHistory: any[] = []) {
    const messages = [
      ...conversationHistory,
      {
        role: "user",
        content: userMessage,
      },
    ];

    let response = await this.openai.chat.completions.create({
      model: "gpt-4o", // or "gpt-4-turbo", "gpt-3.5-turbo"
      messages: messages,
      tools: this.availableTools,
      tool_choice: "auto",
    });

    let message = response.choices[0].message;
    console.log(`Response finish reason: ${response.choices[0].finish_reason}`);

    // Handle tool calls iteratively
    while (message.tool_calls && message.tool_calls.length > 0) {
      // Add assistant's response to conversation
      messages.push(message);

      // Execute each tool call
      for (const toolCall of message.tool_calls) {
        const toolName = toolCall.function.name;
        const toolArgs = JSON.parse(toolCall.function.arguments);

        // Execute the tool via Zyfai MCP
        const toolResult = await this.executeTool(toolName, toolArgs);

        // Add tool result to conversation
        messages.push({
          role: "tool",
          tool_call_id: toolCall.id,
          content: toolResult,
        });
      }

      // Get next response from OpenAI
      response = await this.openai.chat.completions.create({
        model: "gpt-4o",
        messages: messages,
        tools: this.availableTools,
        tool_choice: "auto",
      });

      message = response.choices[0].message;
      console.log(
        `Continued finish reason: ${response.choices[0].finish_reason}`
      );
    }

    // Extract final text response
    const finalResponse = message.content || "";

    return {
      response: finalResponse,
      conversationHistory: [...messages, message],
    };
  }
}

/**
 * Example Usage: AI-Powered DeFi Scenarios
 */
async function runDeFiAgent() {
  // Initialize Zyfai MCP client
  const zyfaiClient = await initializeZyfaiMCP();

  // Create AI agent with Zyfai tools (using OpenAI)
  const agent = new DeFiAIAgent(zyfaiClient, process.env.OPENAI_API_KEY!);
  await agent.initialize();

  // Scenario 1: Find best yield opportunities
  console.log("\nScenario 1: Finding best yields\n");
  const result1 = await agent.chat(
    "I have $10,000 USDC on Base chain. Find me the safest yield opportunities with at least 4% APY. Compare the top 3 options and recommend the best one considering TVL and historical performance."
  );
  console.log("Agent:", result1.response);

  // Scenario 2: Portfolio analysis with multi-chain context
  console.log("\nScenario 2: Portfolio Analysis\n");
  const result2 = await agent.chat(
    "Analyze the portfolio for wallet 0x742d35Cc6634C0532925a3b844Bc9e7595f0bEb across Base and Arbitrum. Calculate total earnings, identify underperforming positions, and suggest rebalancing strategies.",
    result1.conversationHistory
  );
  console.log("Agent:", result2.response);
}

// Run examples
console.log("Starting AI DeFi Agent Examples...\n");

runDeFiAgent()
  .then(() => {
    console.log("\nAll scenarios completed!");
  })
  .catch((error) => {
    console.error("Error:", error);
  });

Example Conversation:

User: "I have 5000 USDC on Base. What's my best option?"

Agent: *Calls get-conservative-opportunities tool*
       *Calls get-available-protocols tool*
       *Analyzes results*

Agent: "Based on current data, I recommend Aave V3 on Base:
       - APY: 4.8% (stable over 30 days)
       - TVL: $45M (high liquidity)
       - Risk: Low (audited, established protocol)
       - Expected monthly return: ~$20

       Alternative: Compound offers 4.2% APY if you prefer even
       more battle-tested protocols."

User: "How has Aave performed historically?"

Agent: *Calls get-daily-apy-history tool*
       *Calls get-apy-per-strategy tool*

Agent: "Aave on Base has shown consistent performance:
       - 30-day average APY: 4.6%
       - Lowest: 4.1%, Highest: 5.2%
       - Trend: Stable with slight upward movement
       - Active wallets: 2,847 (growing ecosystem)

       This is a solid choice for stable, predictable yields."

For more advanced integrations, combine with direct @zyfai/sdk integration for transaction execution

Getting Started

Prerequisites

• Node.js >= 18.0.0
• pnpm (or npm)
• Zyfai API Key (optional, for authenticated endpoints)

Local Development

1. Clone the repository:

git clone https://github.com/ondefy/zyfai-mcp-server
cd zyfai-mcp-server

2. Install dependencies:

pnpm install

3. Create environment file (optional):

# Create .env file
cat > .env <<EOF
PORT=3005
HOST=0.0.0.0
ALLOWED_ORIGINS=*
ZYFAI_API_KEY=your_api_key_here
EOF

4. Configure environment variables:

Get your ZYFAI_API_KEY from Zyfai SDK Dashboard

Note: The server will start without an API key (shows a warning), but API calls will fail. Set ZYFAI_API_KEY in your .env file for full functionality.

5. Build the project:

pnpm run build

6. Start the server:

pnpm start

The server will be running at:

• Main endpoint: http://localhost:3005/
• MCP endpoint: http://localhost:3005/mcp
• Health check: http://localhost:3005/health

Development Mode

pnpm run dev

This will build and start the server in development mode.

Environment Variables

Configure your server using environment variables:

Variable	Description	Default	Required
PORT	Server port	3005	No
HOST	Host to bind to	0.0.0.0	No
ZYFAI_API_KEY	Zyfai SDK API key	-	Yes (for API calls)
ALLOWED_ORIGINS	CORS allowed origins (comma-separated)	*	No

Available Scripts

• pnpm run build - Compile TypeScript to JavaScript
• pnpm start - Start the production Streamable HTTP server (build/index.js)
• pnpm run start:stdio - Start the STDIO server for Claude Desktop (build/index-stdio.js)
• pnpm run dev - Build and start in development mode
• pnpm run clean - Clean build directory

Note: The project includes:

• index.ts - Streamable HTTP server for web/remote access
• index-stdio.ts - STDIO server for Claude Desktop (local)
• proxy-server.ts - Proxy server bridging stdio to remote Streamable HTTP

Migration from SSE (Legacy)

If you're upgrading from a previous version using SSE transport:

1. Update @modelcontextprotocol/sdk to ^1.12.0 or later
2. Change client transport from SSEClientTransport to StreamableHTTPClientTransport
3. Update endpoint URLs from /sse to /mcp
4. Add Mcp-Session-Id header handling (automatically managed by the SDK)

Before (SSE):

import { SSEClientTransport } from "@modelcontextprotocol/sdk/client/sse.js";
const transport = new SSEClientTransport(new URL("https://mcp.zyf.ai/sse"));

After (Streamable HTTP):

import { StreamableHTTPClientTransport } from "@modelcontextprotocol/sdk/client/streamableHttp.js";
const transport = new StreamableHTTPClientTransport(
  new URL("https://mcp.zyf.ai/mcp")
);

Contributing

Feel free to submit issues and enhancement requests!

License

ISC

Related Projects

• @zyfai/sdk - Zyfai TypeScript SDK
• @modelcontextprotocol/sdk - Model Context Protocol SDK