Brief Summary
This video provides a comprehensive guide on how to build a voice agent from scratch using Claude Code and Vapi's new skills. It covers setting up the Vapi MCP, installing the necessary skills, and building a voice agent capable of booking appointments and logging call data. The video also highlights the importance of skills in simplifying the development process and the potential for AI to automate business tasks.
- Vapi skills simplify voice agent creation.
- Claude Code can automate most of the building process.
- AI agents require verification and occasional manual adjustments.
Intro
Vapi has simplified the creation of voice agents with the release of official skills for Claude Code. This allows users to build voice agents from scratch without needing to use the Vapi dashboard or read extensive documentation. The process involves instructing Claude Code on the desired functionality, and it handles the build. The video will demonstrate how to set up the Vapi MCP, install the eight newly introduced skills, and build a voice agent within Claude Code.
Full Demo + Results
The video demonstrates a call with a voice AI receptionist built using Claude Code. The AI, named Rachel, books an appointment for a deck construction estimate. The demo shows Rachel asking relevant questions about location, timeline, and budget, and then scheduling an on-site estimate. The results of the call are then reviewed, showing the appointment correctly added to Google Calendar and the call data logged in a CRM, although with a minor misspelling. An end-of-call report is also sent via email, including the caller's information, call cost, full transcript, and a link to the call recording.
What Are Vapi Skills?
Vapi skills are knowledge files that teach an AI coding assistant how to use specific tools. Vapi has introduced its skills and MCP (Multi-Call Processing), which includes eight skills such as setting up an API key, creating an assistant or tool, creating a call or squad, creating a phone number, and setting up a web hook. The MCP connector can query documents for information beyond these skills. The MCP and skills are compatible across different agents like Claude Code and Gemini CLI.
Installing Vapi MCP & Skills
To install the Vapi MCP and skills, the presenter recommends copying the entire blog post with instructions into Claude Code or another coding platform and instructing it to follow the instructions. This involves obtaining the API key from Vapi and allowing Claude Code to handle the setup. Once installed, the skills can be verified by using commands like /create to see available skills or /mcp status to check the MCP server status.
Phase 1
The first phase involves setting up the project with a claw.md file, which provides an overview of the project and instructions. The initial prompt, "Let's build a voice agent," triggers Claude Code to guide the user through setting up the Vapi API key. Claude Code identifies the missing Vapi API key and prompts the user to run the Vapi setup.
Phase 2
To begin phase two, the command "start phase two" is used. Due to security measures to protect environment variables, the API keys are re-entered from the .env file. Claude Code then creates tools such as Google Calendar check availability, Google Calendar create event, and log lead tool. These tool configurations are saved to a tool_configs.json file, and the call and transfer functions are enabled.
Phase 3
Phase three involves creating the AI assistant. Claude Code creates an assistant named Rachel for BuildRight Receptionist, using the OpenAI GPT-4.0 model and the Vapi Clara voice. The system prompt is generated, and the assistant is configured with a maximum token count of 300 and a temperature of 0.7. Inline tools are connected, and the end call function is enabled.
Phase 4
In phase four, Claude Code creates a free Vapi phone number and assigns it to Rachel, the assistant. This allows for both inbound and outbound calls. The phone number is created and linked to the BuildRight Construction mainline, enabling Rachel to handle calls.
Phase 5 & 6
Phase five focuses on creating a web hook and an N&N (N8N) workflow to capture the end-of-call report and send an email. The server URL for Vapi is created and connected, and the N&N workflow is imported. Phase six involves testing the setup. An outbound test call is initiated using the Vapi create call tool, and an inbound test call is performed to ensure the N&N workflow functions correctly.
AI Can't One-Shot Everything
The presenter emphasises that AI cannot automate everything perfectly. In this case, the initial N&N end workflow was not set up correctly and required troubleshooting. After some adjustments and testing, the workflow was fixed. The remaining tests, including an inbound test call and verification of the email notification, were then completed successfully.
Join AIS Plus
The presenter recommends joining the AI Automation Society Plus community to learn how to make money with AI and begin an AI journey. The community offers courses on mastering N8N, learning about cloud code, and running an AI automation agency. It also provides access to a community of over 3,000 members actively making money with AI.
My 3 Biggest Takeaways
The presenter shares three key takeaways from working with Vapi skills:
- Skills significantly improve Vapi builds, reducing the process from hours of dashboard work to 10-20 minutes in Cloud Code.
- Vapi has effectively packaged its platform into skills, allowing Claude to build with it natively.
- The agent skills standard is becoming universal, with companies like Stripe and Notion adopting it, making AI assistants experts in various software tools.
Outro
The presenter encourages viewers to like the video, subscribe to the channel, and leave a comment about the type of voice agent they plan to build.
Blessing
The presenter concludes the video with a prayer, asking for protection and blessings for the viewers, their businesses, and their families.
