It used to take me a couple hours to configure a voice agent. The knowledge base. The voice. The prompt. The phone number. All of it.
Yesterday, I used Claude Code to do it in one terminal command.
Three voice agents. Five minutes each. All from the terminal. No platform-hopping. No manual configuration. Just plain English and an MCP server.
I’m Charles Dove, and at Charlie Automates we strip away the hype and show you what actually works. This is one of those things.
What You Need (The Full Stack)
Here’s the thing. The setup is dead simple. Three pieces:
- Claude Code (the terminal, not the web app)
- ElevenLabs MCP Server (the repo link is in the video description)
- A Twilio phone number (free tier works to start)
That’s it. No complicated integrations. No custom code. Claude Code handles the heavy lifting through the MCP server.
Step 1: Install the ElevenLabs MCP Server
Go to the ElevenLabs MCP server repo. Open a session of Claude Code. Paste the repo link and ask Claude to set it up for you.
Claude will configure everything. It’ll ask you for your ElevenLabs API key.
Head to 11labs.io. Create an account. You can go free to start, or do the $20/month membership. It’s really not that expensive.
Navigate to the developer section. Create your API key. You can leave it unrestricted since you won’t be sharing it with anyone.
Take that API key back to Claude Code. Plug it in. Claude hooks it up with the MCP server automatically.
That’s the technical setup. You’re ready to go.
Step 2: Plan Your Agents in Plan Mode
Here’s where it gets interesting. As good practice, hit Shift+Tab to switch to Plan Mode. This tells Claude to ask you questions before building anything.
I created three separate agents to show the variance:
- Jessica - Intake receptionist for a personal injury law firm
- Amy - Customer support agent for a construction company
- Rachel - After hours agent for an accounting firm
I have a custom “create agent” skill in my workspace. If you want access to the markdown file, you can grab it from my Skool community. But you don’t need it to follow along.
How Claude Asks the Right Questions
Claude is super intelligent with these conversations. For the law firm agent, it asked about:
- The agent’s name (Jessica)
- Practice area (personal injury)
- Primary function (collect info and schedule consultations)
- Tone (warm and empathetic)
- Voice gender preference
For the construction company, I went with friendly and patient. For the accounting firm, professional and efficient. Claude picks up on these differences and configures everything accordingly.
If you were to go on the ElevenLabs website yourself, you’d have to learn the whole platform. With the MCP server, Claude translates your plain English into the right settings. It’s not just saving you time. It’s more effectively putting the configuration together based on the conversation you’re having.
Step 3: Add Knowledge Bases
With this MCP server, you get access to add data to each agent’s knowledge base. You can add PDFs, documents, or just plain text.
I instructed Claude to create fake knowledge bases for all three agents. Company addresses, founding year, owner names, and FAQs. This way each agent actually knows about the business it represents.
Claude added all three knowledge bases in parallel. One for each agent. The “create agent” tool handles one part. The knowledge base tool handles another. Each MCP server has different tools in the toolbox.
Pro tip from my work at CC Strategic: When creating multiple agents or a plan like this, I always recommend creating a custom markdown file you can reference. That way you can clear the context window and execute the plan in a fresh session. But with the Opus 4.6 drop (which I’m in love with), we get a million tokens per context window. Up from 200k with 4.5. Five times the context. Game changer.
Testing the Agents
Once Claude finished building, I tested each one. Here’s what happened:
Jessica (Law Firm): “Hi, thank you for calling Casegate Law. My name is Jessica. I’m so sorry you’re going through this. I’m here to help. Can you tell me a little about what happened?”
She asked about the situation, confirmed safety, gathered info for the attorney. Warm. Empathetic. Exactly what you’d want for a personal injury intake.
Amy (Construction): “Hi there. Thanks for calling ABC Construction. I’m Amy. How can I help you today?”
She handled a frustrated caller perfectly. Asked for name, contact info, and project details. Friendly and patient, just like I specified.
Rachel (Accounting): “Thank you for calling Smith and Associates CPA. You’ve reached our after hours line. My name is Rachel.”
When I asked about taxes and fixing books, she redirected properly: “That’s a great question regarding taxes, but I want to ensure you get the right answers from one of our CPAs.”
Each agent had different voices, different system prompts, and different personalities. All from one conversation in the terminal.
Activating Your Agent with a Phone Number
To actually activate these agents, you need to attach a phone number. I use Twilio for this.
Getting a Twilio number is simple. Start for free on their site. If you don’t understand the process, you can use the Claude Chrome extension. Type “/chrome” inside Claude in the terminal. It’ll activate the browser and walk you through setting up your Twilio number step by step.
Once you have the number, you import it into ElevenLabs. Add the SID from Twilio, assign it to your agent. Done.
You can set it up for inbound calls, outbound calls, or both. This works for clients. For your own business. For whatever purpose you want.
3 Ways to Make Money with Voice Agents
Here’s the thing. Building voice agents is cool. But let’s talk about how to actually make money with them. There are three solid ways.
Way #1: Use Them in Your Own Business
If you’re a business owner and you clock out at 5pm, you’re home with the kids. You don’t want to be taking sales calls or support inquiries from 5pm until 9am the next day.
Set up a voice agent. Position it on your site. When calls come in after hours, they route directly to the AI agent.
Next morning at 9am, check your calendar. See the new appointments that came in overnight. See the support inquiries. Tackle them fresh.
People just want to be heard. If you’re not available, these voice agents make you available even when you’re not around. That alone saves you from missing revenue.
Way #2: Build a SaaS Product (AI as a Service)
Instead of building a voice agent for one client, build it once and sell it to a hundred.
SaaS stands for software as a service. You create the product, charge an upfront fee, then add a monthly recurring structure.
This works perfectly when you niche down. Find a specific type of business you want to service. Construction companies. Law firms. Dental offices. IT companies. Staffing agencies. Pick your niche and build the product.
But here’s what matters: You have to validate the market first. This is from my 12 years of business experience, not just tech hype. Find a legit problem you can solve for a specific niche. Talk to professionals in that space before you build anything.
Don’t spend all this time building a no-code app with voice agents that nobody actually wants to buy. Get product-market fit. PMF. That’s the real work.
Productize the demo. Create a simple landing page where prospects can try the voice agent themselves. The agent asks them questions about their company, what they need, how it would work for their situation. They roleplay with it. They see the value.
Then a simple call-to-action button at the bottom. Book a call. The demo already did the selling. You’re just taking orders at that point.
Way #3: Build a Voice-Powered App
There are plenty of ways to create voice-powered apps. One idea I heard about: an AI agent on a landing page teaching people how to speak different languages.
One of the coolest things about ElevenLabs is that you can train it on multiple languages. Have a dropdown for Chinese and English. Somebody wants to learn Chinese? Your agent teaches them through conversation.
That’s just one example. The key is always the same: validate the market first. Get product-market fit before you scale.
The Sub-Agent Playbook
As promised in the video, I shared my sub-agent guide as a Google doc in the description. Sub-agents are used naturally in Claude Code, but I’ve trained my claude.md file to use them in very specific ways.
This is the kind of deeper content we get into inside the CC Strategic AI community on Skool. If you want more information on how to actually build a full SaaS or AI-as-a-service system, that’s where you’ll find it.
Why This Matters for Your Business
Let me put this in perspective. Voice agents used to require:
- Learning a new platform
- Configuring voices manually
- Writing system prompts from scratch
- Setting up knowledge bases one at a time
- Connecting phone numbers through multiple dashboards
Now it’s one conversation in Claude Code.
At Charlie Automates and our agency CC Strategic, we build these systems for clients every week. The speed difference is real. What used to take hours now takes minutes. And the quality is actually better because Claude understands context and configures settings that make sense for your use case.
Ready to Start Building?
If you want to go deeper on Claude Code, voice agents, and AI automation for your business, here’s where to go next:
Learn with a community: Join CC Strategic AI on Skool for templates, guides, and direct access to what we’re building.
Work with me directly: If you want hands-on help building voice agents, AI systems, or automation for your business, book a 1-on-1 coaching session.
Need it done for you: Book a call with CC Strategic and let our agency handle the build.
Subscribe on YouTube: Follow @charlieautomates for weekly tutorials on Claude Code, AI automation, and building systems that actually work.
FAQ
Do I need a paid ElevenLabs account to build voice agents?
You can start with the free tier. The $20/month plan gives you more minutes and features, but you can test everything on free first.
How much does a Twilio phone number cost?
Twilio lets you start for free. Phone numbers run about $1/month plus per-minute charges. It’s one of the cheapest ways to get a real phone number for your agent.
Can I use Claude Code to build voice agents without coding?
Yes. That’s the whole point. You’re having a conversation in the terminal. Claude handles all the technical configuration through the MCP server. You just describe what you want in plain English.
What’s the difference between Plan Mode and regular mode in Claude Code?
Plan Mode (Shift+Tab) tells Claude to ask you questions and plan before taking action. Regular mode executes immediately. For voice agents, Plan Mode is better because you want to discuss the agent’s personality, tone, and purpose before Claude builds it.
Can these voice agents book appointments?
Yes. ElevenLabs supports agent tools for booking appointments, checking calendars, and sending texts. The MCP server doesn’t support hooking up those tools yet, but you can configure them directly in ElevenLabs.
How do I add a knowledge base to my voice agent?
Through the MCP server, you can add data as PDFs, documents, or plain text. Claude creates the knowledge base and attaches it to your agent in the same conversation. Include company info, FAQs, pricing, and anything else your agent needs to reference.
Can voice agents make outbound calls too?
Yes. You can set up your agent for inbound calls, outbound calls, or both. Outbound is great for follow-ups, appointment reminders, or proactive outreach.
What’s the best niche for selling voice agents?
Any business that takes phone calls and has after-hours gaps. Law firms, dental offices, construction companies, accounting firms, staffing agencies, IT companies. Pick one niche, validate the market, build the product, then scale.
What is an MCP server?
MCP stands for Model Context Protocol. It’s a way for Claude Code to connect with external tools and services. The ElevenLabs MCP server gives Claude direct access to create agents, configure voices, add knowledge bases, and manage phone numbers, all from the terminal.
How long does it really take to set up a voice agent this way?
About 5 minutes per agent once you have the MCP server installed and your API keys ready. I built three agents in the video in under 10 minutes total. Your first one might take a bit longer as you learn the flow, but it gets faster every time.