Setting Up a Claude Agent with MCP

Overview

This playbook walks through integrating Anthropic's Model Context Protocol (MCP) into a Claude-powered agent, enabling it to interact with external tools and data sources via a standardized interface. By the end, your agent will be able to call MCP servers for file access, database queries, and API interactions.

Prerequisites

• Node.js 18+ (or Bun 1.x)
• Anthropic API key with Claude 3.5 or higher access
• An MCP-compatible tool server (e.g., filesystem, postgres, or custom)
• TypeScript 5.x recommended

Step-by-Step Implementation

1. Install the SDK and MCP client

   Add @anthropic-ai/sdk and @modelcontextprotocol/sdk to your project.

2. Configure your MCP server connection

   import { Client } from '@modelcontextprotocol/sdk/client/index.js';
   import { StdioClientTransport } from '@modelcontextprotocol/sdk/client/stdio.js';
   
   const transport = new StdioClientTransport({
     command: 'npx',
     args: ['-y', '@modelcontextprotocol/server-filesystem', './data'],
   });
   const client = new Client({ name: 'my-agent', version: '1.0.0' }, {});
   await client.connect(transport);

3. Wire up tool use in your Claude call

   Fetch available tools from the MCP client, pass them in the tools array to messages.create(), and handle tool_use blocks in the response by routing them back to the MCP client.

Key Considerations

• Always validate tool inputs before passing them to MCP — treat MCP servers as untrusted external services.
• Use max_tokens budgets carefully; tool-heavy agents can exhaust context quickly with large responses.

Next Steps

Once your agent is running, add a retry loop for tool_use blocks and consider a prompt caching strategy to reduce latency on repeated system prompts. See the Anthropic Cookbook for production-grade agent patterns.