Unified MCP Client Library

🌐 MCP Client is the open-source way to connect any LLM to any MCP server in TypeScript/Node.js, letting you build custom agents with tool access without closed-source dependencies.

💡 Let developers easily connect any LLM via LangChain.js to tools like web browsing, file operations, 3D modeling, and more.

---

✨ Key Features

Feature	Description
🔄 Ease of use	Create an MCP-capable agent in just a few lines of TypeScript.
🤖 LLM Flexibility	Works with any LangChain.js-supported LLM that supports tool calling.
🌐 HTTP Support	Direct SSE/HTTP connection to MCP servers.
⚙️ Dynamic Server Selection	Agents select the right MCP server from a pool on the fly.
🧩 Multi-Server Support	Use multiple MCP servers in one agent.
🛡️ Tool Restrictions	Restrict unsafe tools like filesystem or network.
🔧 Custom Agents	Build your own agents with LangChain.js adapter or implement new adapters.

---

🚀 Quick Start

Requirements

• Node.js 22.0.0 or higher
• npm, yarn, or pnpm (examples use pnpm)

Installation

# Install from npm
npm install mcp-use
# LangChain.js and your LLM provider (e.g., OpenAI)
npm install langchain @langchain/openai dotenv

Create a .env:

OPENAI_API_KEY=your_api_key

Basic Usage

import { ChatOpenAI } from '@langchain/openai'
import { MCPAgent, MCPClient } from 'mcp-use'
import 'dotenv/config'

async function main() {
  // 1. Configure MCP servers
  const config = {
    mcpServers: {
      playwright: { command: 'npx', args: ['@playwright/mcp@latest'] }
    }
  }
  const client = MCPClient.fromDict(config)

  // 2. Create LLM
  const llm = new ChatOpenAI({ modelName: 'gpt-4o' })

  // 3. Instantiate agent
  const agent = new MCPAgent({ llm, client, maxSteps: 20 })

  // 4. Run query
  const result = await agent.run('Find the best restaurant in Tokyo using Google Search')
  console.log('Result:', result)
}

main().catch(console.error)

---

📂 Configuration File

You can store servers in a JSON file:

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["@playwright/mcp@latest"]
    }
  }
}

Load it:

import { MCPClient } from 'mcp-use'
const client = MCPClient.fromConfigFile('./mcp-config.json')

---

🔄 Multi-Server Example

const config = {
  mcpServers: {
    airbnb: { command: 'npx', args: ['@openbnb/mcp-server-airbnb'] },
    playwright: { command: 'npx', args: ['@playwright/mcp@latest'] }
  }
}
const client = MCPClient.fromDict(config)
const agent = new MCPAgent({ llm, client, useServerManager: true })
await agent.run('Search Airbnb in Barcelona, then Google restaurants nearby')

---

🔒 Tool Access Control

const agent = new MCPAgent({
  llm,
  client,
  disallowedTools: ['file_system', 'network']
})

👥 Contributors

Zane

📜 License

MIT © Zane