Order MCP

A small end-to-end example of a Groq-powered order assistant using the Model Context Protocol (MCP).

Architecture

Groq LLM -> MCP client -> MCP server -> REST order backend

• src/ai-client.ts sends the user question to Groq and acts as the MCP client.

• src/index.ts is the MCP server and exposes the get_order_status tool over stdio.

• src/backend.ts is a sample REST order backend at http://localhost:3000.

Related MCP server: woocommerce-mcp-server

Tool discovery and calling

After connecting to the MCP server, the client calls listTools() to discover the tools and their JSON schemas. It converts those definitions to Groq function tools and includes them in the chat request. When Groq returns a tool call, the client passes its name and arguments to callTool(). The MCP server calls the REST backend, and the client sends the tool result back to Groq for the final response.

Install and run

Requires Node.js 20 or later.

npm install

Supply a Groq API key through the GROQ_API_KEY environment variable. Do not put the key in source code or commit it to Git.

Start the REST backend in one terminal:

npm run backend

Then run the client in another terminal. The client starts and connects to the MCP server automatically:

PowerShell:

$env:GROQ_API_KEY = "your-key"
npm run client

macOS/Linux:

export GROQ_API_KEY="your-key"
npm run client