Let me share a concrete example of a multi-agent setup using MCP tools and full data sharing — the Restaurant Agent system. Here’s a friendly walkthrough of how it works: • The Greeter Agent warmly welcomes the user and figures out if they want to make a reservation or place a takeaway order. It gently gathers initial details like customer preferences and smoothly passes this information to the right specialized agents. • Depending on what the user wants, control is handed off to either the Reservation Agent or the Takeaway Agent. These agents share all the collected data through a common

UserData

object (like customer name and phone number), so nothing needs to be repeated as the conversation moves along. • Then, the Checkout Agent steps in from the Takeaway Agent to handle payments. It has full access to the order history and customer info, making the process easy and seamless without asking the user to re-enter any details. Each agent uses handy function tools (such as

update_order

and

confirm_reservation

) to keep everything consistent and flowing smoothly. On top of that, MCP tools come into play by specifying

mcp_servers

when setting up the

AgentSession

. This opens up cool external capabilities like pulling data from Shopify stores or automating tasks through Zapier. For instance, in the Shopify Voice Shopper demo, an agent uses MCP to fetch product info and send navigation commands to a browser frontend, showing how MCP enhances agent features while keeping the session connected. If you’d like to dive deeper into how this all fits together, here are some helpful resources to explore: • Restaurant Agent example • MCP integration examples • Shopify Voice Shopper with MCP Sources: Workflows | LiveKit Docs | complex-agents/shopify-voice-shopper/README.md | agent.py | examples/src/multi_agent.ts | main.py

Let me share a concrete example of a multi-agent setup using MCP tools with full data sharing — the Restaurant Agent system. It’s designed to make interactions smooth and intuitive for users: • The Greeter Agent warmly welcomes the user and figures out if they want to make a reservation or place a takeaway order. It gathers important details like customer preferences and then passes this information along seamlessly to the right specialized agents. • Depending on what the user wants, control is handed over to either the Reservation Agent or the Takeaway Agent. These agents have access to all previously collected data through a shared

UserData

object — things like the customer’s name and phone number — which stays consistent as the conversation moves along. • The Checkout Agent steps in after the Takeaway Agent to handle payments. It can access the full order history and customer info, so the user doesn’t have to repeat themselves. Each agent uses handy function tools (like

update_order

and

confirm_reservation

) to keep everything up-to-date and running smoothly. On top of that, MCP tools are integrated by specifying

mcp_servers

when you start the

AgentSession

. This opens up external capabilities, such as pulling data from Shopify stores or automating tasks through Zapier. For instance, in the Shopify Voice Shopper demo, an agent uses MCP to fetch product details and even send navigation commands to a browser frontend — showing how MCP can really extend what agents can do while keeping the session context intact. If you’d like to dive deeper into how this all works, here are some great resources to explore: • Restaurant Agent example • MCP integration examples • Shopify Voice Shopper with MCP