@4ourlab/mcp-client-claude

A MCP (Model Context Protocol) implementation for Claude models that allows connecting and using multiple MCP servers through Anthropic's Claude API.

This implementation is based on the official Model Context Protocol documentation.

Features

• 🔗 Connect to multiple MCP servers
• 🤖 Integration with Anthropic Claude API
• 📝 Support for custom system prompts
• 🔧 Complete TypeScript interface
• 🛠️ Included usage examples

Supported Models

This client has been tested with the following Claude models:

• claude-3-5-sonnet-20241022
• claude-sonnet-4-20250514

Installation

npm install @4ourlab/mcp-client-claude

Basic Usage

import { MCPClient } from '@4ourlab/mcp-client-claude';

const mcpClient = new MCPClient(
    "your-anthropic-api-key",
    "claude-3-5-sonnet-20241022", // or any other supported model
    "./path/to/mcpServer.json",
    "Optional system prompt"
);

try {
    await mcpClient.connectToServers();
    const response = await mcpClient.processQuery("Your query here");
    console.log(response);
} finally {
    await mcpClient.cleanup();
}

MCP Server Configuration

Create an mcpServer.json file with your server configuration:

{
    "mcpServers": {
        "weather": {
            "command": "node",
            "args": ["/path/to/mcpserver-weather/build/index.js"]
        },
        "mssql": {
            "command": "dotnet",
            "args": ["run", "--project", "/path/to/mcpserver-mssql.csproj"]
        }
    }
}

Examples

Example 1: Interactive Chat

import { MCPClient } from '@4ourlab/mcp-client-claude';

async function main() {
    const mcpClient = new MCPClient(
        "your-api-key",
        "claude-3-5-sonnet-20241022",
        "./examples/mcpServer.json",
        ""
    );

    try {
        await mcpClient.connectToServers();
        await mcpClient.chatLoop();
    } finally {
        await mcpClient.cleanup();
        process.exit(0);
    }
}

main().catch(console.error);

Example 2: Query Processing with JSON Response

import { MCPClient } from '@4ourlab/mcp-client-claude';

async function main() {
    const systemPrompt = `
        You are an intelligent assistant with access to tools. Use your knowledge and available tools to solve problems proactively. 

        For final responses, use JSON format:
        {
            "header": {
                "success": true|false,
                "usedTools": true|false,
                "message": "error description when success=false"
            },
            "result": {
                "your response content here"
            }
        }`;

    const mcpClient = new MCPClient(
        "your-api-key",
        "claude-3-5-sonnet-20241022",
        "./examples/mcpServer.json",
        systemPrompt
    );

    try {
        await mcpClient.connectToServers();
        const response = await mcpClient.processQuery("What's the weather in Sacramento?");
        console.log("\nResponse:\n" + cleanResponse(response));
    } catch (error) {
        console.error("Error in main:", error);
    } finally {
        await mcpClient.cleanup();
        process.exit(0);
    }
}

function cleanResponse(response) {
    const content = response;
    const json = content.match(/```json\n([\s\S]*?)\n```/)?.[1] || content.match(/\{[\s\S]*\}/)?.[0];
    return json || response;
}

main().catch(console.error);

API

MCPClient

Constructor

new MCPClient(apiKey: string, model: string, serverConfigPath: string, systemPrompt?: string)

Methods

• connectToServers(): Connect to all configured MCP servers
• processQuery(query: string): Process a query using available servers
• chatLoop(): Start an interactive chat loop
• cleanup(): Clean up connections and resources

Dependencies

• @anthropic-ai/sdk: Official Anthropic Claude API client
• @modelcontextprotocol/sdk: Official MCP SDK

Development

Building the project

npm run build

Development mode

npm run dev

License

MIT