Claude Code Voice

Voice conversations with Claude Opus 4.5 about your code.

Talk through problems, brainstorm ideas, or get a code review — all over the phone.

Quick Start

# Install
pip install claude-code-voice

# One-time setup
claude-code-voice setup

# Register a project
cd your-project
claude-code-voice register

# Start everything (server + tunnel + config)
claude-code-voice start

That's it! Now you can:

• Have Claude call you: claude-code-voice call "debug the auth flow"
• Call Claude: Dial your Vapi number and Claude answers with your project loaded

What You Need

Before running setup:

1. Vapi account — https://vapi.ai (free to sign up, $10 credit included)
2. Vapi API key — https://dashboard.vapi.ai/api-keys
3. Vapi phone number — https://dashboard.vapi.ai/phone-numbers (~$2/month)
4. Node.js — for localtunnel

Note: New Vapi accounts get $10 free credit — enough to test everything without spending anything!

Features

Feature	Description
Opus 4.5	Best-in-class reasoning for technical discussions
Project context	Git status, recent files, todos loaded automatically
Live tools	Claude reads files and searches code during calls
Auto-transcripts	Every call saved as markdown
Personalized	Claude greets you by name
Inbound calls	Call your number and Claude answers

Commands

claude-code-voice setup              # Configure API key, phone, name
claude-code-voice register           # Register current project
claude-code-voice start              # Start server + tunnel (recommended)
claude-code-voice call [topic]       # Have Claude call you
claude-code-voice status             # Check configuration
claude-code-voice config name <name> # Update your name
claude-code-voice config show        # Show all config
claude-code-voice history            # View past calls

Customization

Change voice, model, or language:

# Voice (provider:voiceId)
claude-code-voice config voice openai:nova          # OpenAI Nova
claude-code-voice config voice elevenlabs:rachel    # ElevenLabs Rachel
claude-code-voice config voice deepgram:asteria     # Deepgram Asteria

# Model
claude-code-voice config model claude-sonnet-4-20250514   # Use Sonnet 4

# Language
claude-code-voice config language es    # Spanish
claude-code-voice config language fr    # French

Supported voice providers: openai, elevenlabs, deepgram, playht, azure

How It Works

You ──call──▶ Vapi Phone ──webhook──▶ Your Server ──context──▶ Claude Opus 4.5
                                            │
                                      reads your code

1. Setup stores your Vapi credentials and creates tools
2. Register snapshots project context (git, files, todos)
3. Start runs server + tunnel, auto-configures Vapi
4. Call — Claude has full context about your project

The start Command

The start command is the easiest way to get everything running:

claude-code-voice start

It automatically:

• Kills any existing server on port 8765
• Starts the context server
• Starts localtunnel and waits for URL
• Updates all Vapi tools with the new URL
• Configures inbound call webhooks
• Shows your Vapi number for inbound calls

Just keep that terminal open and you're ready for calls.

Transcripts

Transcripts auto-save to ~/.claude/skills/call/data/transcripts/ when calls end.

Each transcript includes:

• Full conversation
• AI-generated summary
• Call metadata (duration, project, topic)

Troubleshooting

"Connection issues" or "404 errors"

Run claude-code-voice start — it auto-configures everything when the tunnel URL changes.

"I don't recognize this number"

Call from the phone number you used during setup.

Claude can't read files during call

Make sure claude-code-voice start is running in a terminal.

Check your configuration

claude-code-voice status

As a Claude Code Skill

For /call directly in Claude Code:

git clone https://github.com/abracadabra50/claude-code-voice-skill.git
ln -s /path/to/claude-code-voice-skill ~/.claude/skills/call

Then use /call in conversations.

Manual Setup (Advanced)

If you prefer manual control over the server and tunnel:

# Terminal 1: Start server
claude-code-voice server

# Terminal 2: Start tunnel
npx localtunnel --port 8765

# Terminal 3: Configure (run after each tunnel restart)
claude-code-voice config server-url https://xxx.loca.lt
claude-code-voice configure-inbound

License

MIT