How To Build AI Voice Agents Using Claude, Retell AI, And Zapier MCP

Updated: June 5, 2026

This guide explains how to build a fully automated AI calling system where Claude can automatically call customers, handle conversations, store transcripts, and give you summaries without touching your phone manually. The complete system uses Claude as the brain, Retell AI for voice calling, and Zapier MCP to connect everything together.

The final setup allows you to automate tasks like lead follow ups, renewal reminders, feedback collection, customer support calls, and many other business workflows using AI voice agents.

Table of Contents

What This System Actually Does

The workflow is extremely simple from the user side.

You type a single message inside Claude such as:

“Call David and ask if he is still interested in the insurance policy.”

Claude then automatically understands:

• what type of call needs to happen

• which voice agent should be used

• what the objective of the call is

After that, the system:

• places the call automatically

• talks with the customer

• handles the conversation naturally

• stores the transcript

• generates a summary

• sends the result back to you

The entire process happens without manually opening your phone or dialing numbers yourself.

Core Architecture Of The System

The setup shown in the transcript uses three main tools working together.

Claude

Claude acts as the central brain of the system. It receives your instructions and decides which AI voice agent should handle the task.

Claude is responsible for:

• understanding prompts

• selecting the correct agent

• triggering calls

• retrieving summaries

• analyzing transcripts

Retell AI

Retell AI is responsible for the actual phone calls and voice conversations.

It handles:

• voice generation

• AI conversations

• speech recognition

• call management

• multilingual support

Each Retell AI agent is designed for one specialized task.

Zapier MCP

Zapier MCP acts as the bridge between Claude and Retell AI because Retell AI does not directly integrate into Claude by default.

Zapier MCP allows Claude to:

• make phone calls through Retell AI

• retrieve call information

• access transcripts

• automate workflows

Step 1: Create A Retell AI Account

Start by creating a free account on Retell AI.

Retell AI uses a pay as you go pricing model, which means you only pay for the calls you actually make. According to the transcript, even after running multiple tutorials and active agents, the total cost remained around $3.

There is no expensive monthly subscription required to get started.

Step 2: Understand Why Multiple Agents Matter

One of the most important ideas in this system is using multiple specialized agents instead of one general assistant.

Each voice agent should focus on one specific task only.

For example, the transcript showed three different agents:

Lead Follow Up Agent

• Renewal Reminder Agent

• Post Claim Feedback Agent

This structure improves performance because the AI only needs to focus on one workflow at a time.

A specialized agent produces much more reliable and natural conversations compared to one giant universal assistant trying to do everything.

Step 3: Create Your First Voice Agent

Inside Retell AI, click:

Create Agent → Voice Agent

Then choose:

• Single Prompt Agent

• Blank Agent

The transcript specifically recommends starting with a Single Prompt Agent because it keeps the workflow simpler and easier to understand.

More advanced conversational flows can be explored later once the basics are working.

Step 4: Choose The AI Model

Retell AI supports multiple AI models such as:

• GPT 4.1

• Claude models

• Gemini

• GPT 5 Nano

More advanced models usually provide:

• better quality conversations

• stronger reasoning

• more natural interactions

But they also cost more.

The transcript mentioned GPT 5 Nano as a cheaper option, while GPT 4.1 and Claude provide higher quality responses.

Choose the model depending on your balance between quality and cost.

Step 5: Configure The Voice And Language

Retell AI supports many languages including:

• English

• Hindi

• Urdu

• Tamil

• Marathi

• Kannada

The transcript specifically recommends using Hindi if targeting Indian users because the language model has stronger training data and sounds more natural.

You can also customize:

• male or female voice

• accents

• speaking style

Before continuing, test the voice using the “Run Test” feature to make sure the output sounds natural enough for your use case.

Step 6: Write The Agent Prompt

The prompt defines the entire behavior of the AI voice agent.

For example, a Lead Follow Up Agent prompt might instruct the AI to:

• greet the customer politely

• mention the company name

• ask if they are still interested

• answer simple questions

• collect a yes or no response

• end the call professionally

The transcript strongly recommends generating prompts using Claude instead of writing them manually because AI generated prompts often include better structure and logic.

A well written prompt dramatically improves call quality.

Step 7: Configure Agent Settings Properly

Inside Retell AI, several settings are important for realism and performance.

AI First Mode

Enable AI First mode so the AI speaks first when the customer answers the phone.

Without this, conversations can feel awkward or delayed.

Knowledge Base Integration

Retell AI allows you to connect:

• websites

• documents

• PDFs

• product information

For example, if you connect an ecommerce website, the AI can answer customer questions about:

• pricing

• products

• policies

• inventory

• services

directly during the call.

This makes the AI far more useful for real business workflows.

Voice Speed And Interruptions

You can also configure:

• speaking speed

• interruption timing

• pause behavior

The transcript recommends experimenting with these settings until the conversation feels natural and human.

Step 8: Create Multiple Specialized Agents

After your first agent is working, repeat the process for additional workflows.

For example:

Lead Follow Up Agent

Used for checking if leads are still interested.

Renewal Reminder Agent

Used for reminding customers about expiring insurance policies.

Feedback Collection Agent

Used for customer satisfaction and rating calls.

The key advantage is that Claude later chooses the correct agent automatically based on your prompt.

Step 9: Set Up Zapier MCP

Now the system needs to connect Claude with Retell AI.

Go to Zapier MCP and create a free account.

Zapier MCP supports thousands of app integrations and acts as the middleware between Claude and Retell AI.

The transcript explains that the free plan is sufficient for most users unless you exceed around 100 monthly calls.

Step 10: Create An MCP Server

Inside Zapier MCP:

• Create New MCP Server

• Select Claude as the target platform

Then click:

Add Tool

Search for:

Retell AI

This gives Claude access to Retell AI functionality through MCP.

Step 11: Enable The Required Retell AI Tools

Two tools are essential:

Make A Phone Call

This allows Claude to place calls using Retell AI agents.

Get A Call

This allows Claude to retrieve:

• call transcripts

• summaries

• call details

• outcomes

Together, these two tools complete the automation loop.

Step 12: Connect Your Retell AI API Key

Inside Retell AI:

Settings → API Keys

Generate a new API key.

Copy the key and paste it into Zapier MCP when connecting the Retell AI account.

This securely links Retell AI with Claude through MCP.

Step 13: Connect Zapier MCP With Claude

Inside Zapier MCP:

Click “Add To Claude”.

Authorize the connection and restart the Claude desktop application.

Once completed, Claude will now have access to:

• Make Phone Call

• Get Call

directly inside the chat interface.

Step 14: Create The Main Claude System Prompt

Now you need to tell Claude how the system works.

The prompt should explain:

• what agents are available

• what each agent does

• how to choose agents

• what phone number to use

For example:

“You are a voice call assistant with access to Retell AI tools through Zapier MCP. Always select the most appropriate agent based on the user request.”

You also include the Retell AI agent IDs inside this prompt.

Step 15: Get The Agent IDs

Each Retell AI agent has a unique ID.

Inside Retell AI:

Open the agent → Copy Agent ID

Paste those IDs into the Claude system prompt.

This allows Claude to know exactly which voice agent to trigger during a call.

Step 16: Buy A Retell AI Phone Number

To make real calls, you need a phone number inside Retell AI.

Go to:

Phone Numbers → Buy New Number

The transcript mentioned these numbers typically cost around $2 per month.

You can choose:

• US numbers

• Canadian numbers

Once purchased, add this number into your Claude prompt configuration.

Step 17: Test The Full System

At this point, the system is fully operational.

You can now type prompts such as:

“Call David and remind him his insurance policy is expiring.”

Claude will:

• identify the correct agent

• place the call

• conduct the conversation

• retrieve the summary

The transcript demonstrated this exact workflow successfully.

Step 18: Retrieve Call Summaries And Transcripts

After every call, Claude can retrieve:

• full transcripts

• customer responses

• call duration

• conversation summaries

• final outcomes

For example:

“David confirmed he wants to renew the policy.”

This information can later be:

• stored in spreadsheets

• pushed into CRMs

• analyzed with AI

• connected to dashboards

The system is extremely flexible.

Real Business Use Cases

This setup can be used for:

• lead qualification

• appointment reminders

• sales follow ups

• customer support

• insurance renewals

• feedback collection

• AI call centers

• CRM automation

The transcript emphasized that you can scale this setup from:

• 2 agents

• 50 agents

or even

• 100+ specialized agents

Claude automatically handles agent selection for you.

Final Tech Stack

The full stack used in the transcript includes:

Claude for orchestration and reasoning

• Retell AI for voice conversations

• Zapier MCP for integrations

• AI prompts for agent behavior

• Knowledge bases for business information

Final Result

By the end of this setup, you have a fully automated AI voice calling system where Claude can:

• place calls automatically

• talk with customers naturally

• choose specialized agents intelligently

• retrieve summaries instantly

• store transcripts automatically

• scale across multiple workflows

All controlled directly from a single Claude chat window.

The final setup allows you to automate tasks like lead follow ups, renewal reminders, feedback collection, customer support calls, and many other business workflows using AI voice agents.

Table of Contents

What This System Actually Does

The workflow is extremely simple from the user side.

You type a single message inside Claude such as:

“Call David and ask if he is still interested in the insurance policy.”

Claude then automatically understands:

• what type of call needs to happen

• which voice agent should be used

• what the objective of the call is

After that, the system:

• places the call automatically

• talks with the customer

• handles the conversation naturally

• stores the transcript

• generates a summary

• sends the result back to you

The entire process happens without manually opening your phone or dialing numbers yourself.

Core Architecture Of The System

The setup shown in the transcript uses three main tools working together.

Claude

Claude acts as the central brain of the system. It receives your instructions and decides which AI voice agent should handle the task.

Claude is responsible for:

• understanding prompts

• selecting the correct agent

• triggering calls

• retrieving summaries

• analyzing transcripts

Retell AI

Retell AI is responsible for the actual phone calls and voice conversations.

It handles:

• voice generation

• AI conversations

• speech recognition

• call management

• multilingual support

Each Retell AI agent is designed for one specialized task.

Zapier MCP

Zapier MCP acts as the bridge between Claude and Retell AI because Retell AI does not directly integrate into Claude by default.

Zapier MCP allows Claude to:

• make phone calls through Retell AI

• retrieve call information

• access transcripts

• automate workflows

Step 1: Create A Retell AI Account

Start by creating a free account on Retell AI.

There is no expensive monthly subscription required to get started.

Step 2: Understand Why Multiple Agents Matter

One of the most important ideas in this system is using multiple specialized agents instead of one general assistant.

Each voice agent should focus on one specific task only.

For example, the transcript showed three different agents:

Lead Follow Up Agent

• Renewal Reminder Agent

• Post Claim Feedback Agent

This structure improves performance because the AI only needs to focus on one workflow at a time.

A specialized agent produces much more reliable and natural conversations compared to one giant universal assistant trying to do everything.

Step 3: Create Your First Voice Agent

Inside Retell AI, click:

Create Agent → Voice Agent

Then choose:

• Single Prompt Agent

• Blank Agent

The transcript specifically recommends starting with a Single Prompt Agent because it keeps the workflow simpler and easier to understand.

More advanced conversational flows can be explored later once the basics are working.

Step 4: Choose The AI Model

Retell AI supports multiple AI models such as:

• GPT 4.1

• Claude models

• Gemini

• GPT 5 Nano

More advanced models usually provide:

• better quality conversations

• stronger reasoning

• more natural interactions

But they also cost more.

The transcript mentioned GPT 5 Nano as a cheaper option, while GPT 4.1 and Claude provide higher quality responses.

Choose the model depending on your balance between quality and cost.

Step 5: Configure The Voice And Language

Retell AI supports many languages including:

• English

• Hindi

• Urdu

• Tamil

• Marathi

• Kannada

The transcript specifically recommends using Hindi if targeting Indian users because the language model has stronger training data and sounds more natural.

You can also customize:

• male or female voice

• accents

• speaking style

Before continuing, test the voice using the “Run Test” feature to make sure the output sounds natural enough for your use case.

Step 6: Write The Agent Prompt

The prompt defines the entire behavior of the AI voice agent.

For example, a Lead Follow Up Agent prompt might instruct the AI to:

• greet the customer politely

• mention the company name

• ask if they are still interested

• answer simple questions

• collect a yes or no response

• end the call professionally

The transcript strongly recommends generating prompts using Claude instead of writing them manually because AI generated prompts often include better structure and logic.

A well written prompt dramatically improves call quality.

Step 7: Configure Agent Settings Properly

Inside Retell AI, several settings are important for realism and performance.

AI First Mode

Enable AI First mode so the AI speaks first when the customer answers the phone.

Without this, conversations can feel awkward or delayed.

Knowledge Base Integration

Retell AI allows you to connect:

• websites

• documents

• PDFs

• product information

For example, if you connect an ecommerce website, the AI can answer customer questions about:

• pricing

• products

• policies

• inventory

• services

directly during the call.

This makes the AI far more useful for real business workflows.

Voice Speed And Interruptions

You can also configure:

• speaking speed

• interruption timing

• pause behavior

The transcript recommends experimenting with these settings until the conversation feels natural and human.

Step 8: Create Multiple Specialized Agents

After your first agent is working, repeat the process for additional workflows.

For example:

Lead Follow Up Agent

Used for checking if leads are still interested.

Renewal Reminder Agent

Used for reminding customers about expiring insurance policies.

Feedback Collection Agent

Used for customer satisfaction and rating calls.

The key advantage is that Claude later chooses the correct agent automatically based on your prompt.

Step 9: Set Up Zapier MCP

Now the system needs to connect Claude with Retell AI.

Go to Zapier MCP and create a free account.

Zapier MCP supports thousands of app integrations and acts as the middleware between Claude and Retell AI.

The transcript explains that the free plan is sufficient for most users unless you exceed around 100 monthly calls.

Step 10: Create An MCP Server

Inside Zapier MCP:

• Create New MCP Server

• Select Claude as the target platform

Then click:

Add Tool

Search for:

Retell AI

This gives Claude access to Retell AI functionality through MCP.

Step 11: Enable The Required Retell AI Tools

Two tools are essential:

Make A Phone Call

This allows Claude to place calls using Retell AI agents.

Get A Call

This allows Claude to retrieve:

• call transcripts

• summaries

• call details

• outcomes

Together, these two tools complete the automation loop.

Step 12: Connect Your Retell AI API Key

Inside Retell AI:

Settings → API Keys

Generate a new API key.

Copy the key and paste it into Zapier MCP when connecting the Retell AI account.

This securely links Retell AI with Claude through MCP.

Step 13: Connect Zapier MCP With Claude

Inside Zapier MCP:

Click “Add To Claude”.

Authorize the connection and restart the Claude desktop application.

Once completed, Claude will now have access to:

• Make Phone Call

• Get Call

directly inside the chat interface.

Step 14: Create The Main Claude System Prompt

Now you need to tell Claude how the system works.

The prompt should explain:

• what agents are available

• what each agent does

• how to choose agents

• what phone number to use

For example:

“You are a voice call assistant with access to Retell AI tools through Zapier MCP. Always select the most appropriate agent based on the user request.”

You also include the Retell AI agent IDs inside this prompt.

Step 15: Get The Agent IDs

Each Retell AI agent has a unique ID.

Inside Retell AI:

Open the agent → Copy Agent ID

Paste those IDs into the Claude system prompt.

This allows Claude to know exactly which voice agent to trigger during a call.

Step 16: Buy A Retell AI Phone Number

To make real calls, you need a phone number inside Retell AI.

Go to:

Phone Numbers → Buy New Number

The transcript mentioned these numbers typically cost around $2 per month.

You can choose:

• US numbers

• Canadian numbers

Once purchased, add this number into your Claude prompt configuration.

Step 17: Test The Full System

At this point, the system is fully operational.

You can now type prompts such as:

“Call David and remind him his insurance policy is expiring.”

Claude will:

• identify the correct agent

• place the call

• conduct the conversation

• retrieve the summary

The transcript demonstrated this exact workflow successfully.

Step 18: Retrieve Call Summaries And Transcripts

After every call, Claude can retrieve:

• full transcripts

• customer responses

• call duration

• conversation summaries

• final outcomes

For example:

“David confirmed he wants to renew the policy.”

This information can later be:

• stored in spreadsheets

• pushed into CRMs

• analyzed with AI

• connected to dashboards

The system is extremely flexible.

Real Business Use Cases

This setup can be used for:

• lead qualification

• appointment reminders

• sales follow ups

• customer support

• insurance renewals

• feedback collection

• AI call centers

• CRM automation

The transcript emphasized that you can scale this setup from:

• 2 agents

• 50 agents

or even

• 100+ specialized agents

Claude automatically handles agent selection for you.

Final Tech Stack

The full stack used in the transcript includes:

Claude for orchestration and reasoning

• Retell AI for voice conversations

• Zapier MCP for integrations

• AI prompts for agent behavior

• Knowledge bases for business information

Final Result

By the end of this setup, you have a fully automated AI voice calling system where Claude can:

• place calls automatically

• talk with customers naturally

• choose specialized agents intelligently

• retrieve summaries instantly

• store transcripts automatically

• scale across multiple workflows

All controlled directly from a single Claude chat window.

Author

Written By

Vikash Kumar

Building AI agents, n8n workflows and end-to-end automation for 30+ Brands across India, the US, Europe, Dubai & Australia. 7+ years of Experience saving founders real hours every week - no code required.

Ask more Questions about this Blog with AI: