{
  "id": 3059342,
  "title": "Building a restaurant telephony AI host with Amazon Connect",
  "url": "https://urgent.news/2026/08/24/building-a-restaurant-telephony-ai-host-with-amazon-connect",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-24T16:13:09.000Z",
  "source": {
    "name": "AWS Machine Learning",
    "slug": "aws-machine-learning",
    "url": "https://aws.amazon.com/blogs/machine-learning/building-a-restaurant-telephony-ai-host-with-amazon-connect/"
  },
  "original_language": "en",
  "account": "Many restaurants still rely on phone orders, often placing the responsibility of taking those calls on staff already serving customers. The process is inefficient, with callers waiting on hold as orders are taken by hand. Offering an online ordering system helps some customers, but fails to address those who prefer to place orders by phone. This article explains how to construct a voice ordering system that can answer a phone number and take an order, without requiring an app, website, or sign-in. When a caller dials the phone number, an AI host greets them, addresses menu questions, locates the nearest pickup point, and confirms the order verbally. The system employs Amazon Connect for telephony, Amazon Lex V2 combined with Amazon Connect Agentic Voice for real-time speech processing, and Amazon Connect Customer AI agents for managing the conversation, linked to the restaurant's backend via the Model Context Protocol (MCP) and Amazon Bedrock AgentCore. This solution focuses on the telephony channel specifically. Audio is transmitted over the traditional phone network instead of through a web browser, and the system identifies callers by their phone number rather than requiring a login. The solution separates the agent logic and backend services into distinct modules, allowing the ordering logic to remain separate from the channel that calls it. The guide demonstrates how to: Deploy the system using AWS Cloud Development Kit (AWS CDK); Answer incoming phone calls through Amazon Connect and direct them through a contact flow; Utilize Amazon Connect Agentic Voice for real-time speech recognition and synthesis; Manage the conversation and backend tool calls with an Amazon Connect AI agent; Maintain conversation safety and relevance using Amazon Connect AI Guardrail; Connect the agent to backend services as easily discoverable tools via AgentCore Gateway and MCP. Amazon Connect Agentic Voice provides the speech component directly within Amazon Connect, managing both speech recognition and text-to-speech functions. The Amazon Lex V2 bot is responsible for both, while an Amazon Connect AI agent handles reasoning and backend communication. The article also explains how Agentic Voice enhances conversation flow efficiency. The proposed architecture keeps three elements separate: Amazon Connect handles the call, the Amazon Connect AI agent manages the conversation, and the backend stores menu, carts, orders, and locations. When a call arrives through Amazon Connect, a contact flow initiates an Amazon Connect AI agent session, connecting the caller to the agent. Amazon Connect Agentic Voice handles speech recognition and synthesis throughout the call, while the AI agent processes the conversation and communicates with the backend through MCP tools exposed by AgentCore Gateway. Since MCP is an open standard for connecting agents to external tools, the backend can be altered without affecting the agent. In this configuration, the contact flow, Amazon Lex V2 speech layer (Amazon Connect Agentic Voice), and Amazon Connect AI agent are all managed through a single Amazon Connect deployment. The AgentCore Gateway and the restaurant backend are the components you must build and integrate. The solution deploys the following components: Amazon Connect Customer, responsible for inbound telephony, contact flow, and the phone number that accepts calls; Amazon Lex V2, hosting the voice bot connected to the Amazon Connect AI agent; Amazon Connect Agentic Voice, handling speech processing with advanced ASR and expressive TTS natively in Amazon Connect; Amazon Connect Customer AI agents, powered by Anthropic Claude Haiku 4.5 from Amazon Bedrock, orchestrating the conversation; Amazon Connect AI Guardrails, ensuring conversation safety through content filters, denied topics, and profanity filtering; AgentCore Gateway, exposing backend APIs as MCP tools for the agent to discover and call by name; Amazon AppIntegrations, registering the AgentCore Gateway as an MCP application for the AI agent to utilize; Amazon API Gateway, providing REST endpoints secured by AWS Identity and Access Management (IAM); AWS Lambda, executing backend logic for menus, carts, orders, and location lookups; Amazon DynamoDB, storing customer profiles, orders, menu items, carts, and locations; Amazon Location Service, offering geocoding and route calculation for pickup recommendations. Figure 1 illustrates the voice ordering solution, divided into four sections. Section A focuses on backend infrastructure, detailing the deployment of the restaurant backend using Amazon DynamoDB for data storage, Amazon Location Service for address and routing, AWS Lambda for business logic execution, and Amazon API Gateway for REST endpoint management with IAM authentication. Section B outlines the provisioning of an AgentCore Gateway, which imports the REST API's OpenAPI schema during deployment to register each endpoint as a named MCP tool. The gateway enforces custom JSON Web Token (JWT) authorization and validates inbound tokens against the Amazon Connect instance. Section C describes the creation of the Amazon Connect instance, the Amazon Connect AI agent, and the orchestration AI agent. It registers the AgentCore Gateway as an MCP server within Amazon AppIntegrations and establishes an Amazon Lex V2 bot with Amazon Connect Agentic Voice for Advanced ASR. The article then defines the AI Guardrail with safety policies, creates the AI agent with specific configurations, and details the integration process.",
  "summary": "Learn how to build a voice ordering system for restaurants that answers a phone call and takes an order end to end, with no app, no website, and no sign-in. It uses Amazon Connect for telephony, Amazon Connect Agentic Voice for real-time speech, an Amazon Connect AI agent for reasoning, and Amazon Bedrock AgentCore Gateway to reach backend tools through MCP.",
  "key_points": [
    "Amazon Connect AI host takes phone orders without app or website",
    "Amazon Connect Agentic Voice handles speech processing in real-time",
    "Backend services integrated via Model Context Protocol (MCP)"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}