MCP Development in XSky

This skill provides knowledge for Model Context Protocol integrations.

MCP Client Types

Basic Usage

import { SimpleSseMcpClient } from "@xsky/ai-agent-core";

const mcpClient = new SimpleSseMcpClient("http://localhost:3000/mcp");
await mcpClient.connect();

const xsky = new XSky({
  llms: { /* ... */ },
  defaultMcpClient: mcpClient
});

// MCP tools now available to all agents

IMcpClient Interface

interface IMcpClient {
  connect(): Promise<void>;
  disconnect(): Promise<void>;
  listTools(): Promise<McpTool[]>;
  callTool(name: string, args: Record<string, unknown>): Promise<McpToolResult>;
  isConnected(): boolean;
}

Tool Result Format

interface McpToolResult {
  content: Array<{
    type: "text" | "image" | "resource";
    text?: string;
    data?: string;
    mimeType?: string;
  }>;
  isError?: boolean;
}

STDIO Client (Node.js)

import { SimpleStdioMcpClient } from "@xsky/ai-agent-nodejs";

const mcpClient = new SimpleStdioMcpClient("node", ["server.js"]);
await mcpClient.connect();

// Uses JSON-RPC over stdin/stdout

Error Handling

try {
  await mcpClient.connect();
} catch (error) {
  if (error.message.includes('timeout')) {
    // Handle connection timeout
  }
  // Handle other errors
}

Tool Discovery

const tools = await mcpClient.listTools();
// Returns: { name, description, inputSchema }[]

// Tools automatically converted to XSky format
// and registered with agents

Key Source Files