Voice AI agents have come a long way. What used to take a bunch of custom tools and webhook wiring can now be done much faster using MCP. In this video, I show how to build a simple AI receptionist that: - Answers calls - Collects customer info - Checks availability - Books appointments into Google Calendar - Saves the customer record into Google Sheets - Sends a confirmation email Here is the flow, step by step: 1. Set up the voice agent in Vapi - Use GPT 4.1 mini for lower cost and faster responses - Make the assistant speak first - Add a system prompt with identity, tone, and tool instructions 2. Connect Vapi to n8n using MCP - Add an MCP Server Trigger in n8n - Build your tools as sub workflows - In Vapi, create an MCP tool and paste the MCP server URL - Add an Authorization header using your n8n API key 3. Create the sub workflows (these become your tools) - Get user - Search the Google Sheet by email - If not found, return “new client” so the agent knows to create them - Create user - Append a new row with name, phone, email, and call date - Get available slots - Pull calendar events for the requested date - Return busy times between business hours (example 9:00 AM to 5:00 PM) - Book appointment - Create the Google Calendar event - Update the user row with service type and appointment time - Send a confirmation email 4. Let the model fill tool parameters automatically - In the MCP trigger inputs, use “defined automatically by the model” - This avoids extra agents and reduces latency and cost 5. Test the full experience end to end - Call the number - Book an appointment - Confirm the data shows up in Sheets and Calendar - Confirm the email is sent If you want to build this exact setup, I included the resources and workflows in the description. Subscribe for more tutorials on voice AI agents, MCP, and real automations you can deploy. https://youtu.be/IiTV70i-2zA?si=IQAZEllSWNTqoPyH

Today I completed my first workflow on the N8 automation platform , after installing docker container on my modest computer and using the Ngrok program. I installed a version of Llama 3.2 on my computer. Then I set up some simple chat node and simple memory node.. The result was fantastic.