MCP Tools

What is a MCP?

Model Context Protocol (MCP) is an open protocol that standardizes how applications provide context to LLMs. Think of MCP like a USB-C port for AI applications. Just as USB-C provides a standardized way to connect your devices to various peripherals and accessories, MCP provides a standardized way to connect AI models to different data sources and tools.

How to use MCP in Eko?

In Eko, MCP interface extension is provided, the IMcpClient interface is defined as follows:

export interface IMcpClient {
  connect(): Promise<void>;

  listTools(param: McpListToolParam): Promise<McpListToolResult>;

  callTool(param: McpCallToolParam): Promise<ToolResult>;

  isConnected(): boolean;

  close(): Promise<void>;
}

Currently, the MCP interface based on the Server-Sent Events (SSE) request protocol has been implemented in the Eko framework, to use, initialize the SimpleSseMcpClient class.

Usage Example:

import { SimpleSseMcpClient } from "@eko-ai/eko";

let sseUrl = "http://localhost:8080/sse";

let mcpClient = new SimpleSseMcpClient(sseUrl);

Through MCP dynamically expand agent tool

Agent supports dynamic loading of tools through MCP, Assuming you already have an MCP service, The complete example is as follows:

import { Eko, Agent, LLMs, SimpleSseMcpClient } from "@eko-ai/eko";

const llms: LLMs = {
  default: {
    provider: "anthropic",
    model: "claude-3-5-sonnet-20241022",
    apiKey: "your_api_key",
  },
  openai: {
    provider: "openai",
    model: "gpt-4o-mini",
    apiKey: "your_api_key",
  },
};

// your mcp server
let sseUrl = "http://localhost:8080/sse";
let mcpClient = new SimpleSseMcpClient(sseUrl);

// Custom SmartMall agent
let agents: Agent[] = [
  new Agent({
    name: "SmartMall",
    description: "Provide product query and order placement",
    tools: [],
    mcpClient: mcpClient,
    llms: Object.keys(llms),
  }),
];

let eko = new Eko({ llms, agents });
let result = await eko.run("I have $300, please help me buy a set of summer clothes, pants, and shoes.");