Debug code hands-free by voice with Claude Code

Tools

Claude Code

Overview

Watching the OpenClaw/Molbot craze has been entertaining.

Workflow

The author built a local voice interface for Claude Code on an M4 Pro Mac using Parakeet v3 TDT for speech-to-text and Pocket TTS for text-to-speech. A custom MCP server integrates these tools, allowing AirPods stem clicks to interrupt narration and trigger listening. The workflow enables hands-free coding tasks, such as debugging a WhatsApp integration, while the user performs other activities like brushing teeth. Visual output is streamed to a TV via AirPlay to allow remote monitoring of the agent's actions.

Reported result

The author reports the system is extremely fast and functional for hands-free debugging and OS integration tasks. No quantitative metrics on efficiency gains or error rates are provided. The user describes the experience as 'really fun' and beneficial for physical comfort, noting their 'back is thankful.' Success is attributed to the ability to interrupt and guide the agent via voice while multitasking.

Human role

The original post does not say how a person stayed involved.

Limitations

The setup requires specific hardware (M4 Pro, AirPods) and local model execution, which may not be feasible for all users. The author notes the initial implementation was a 'disgusting mess of vibe-coded Python' before migrating to Rust, indicating significant development effort. Security concerns are raised regarding the OpenClaw/Molbot inspiration, though the author's custom stack is local. The system relies on hacky workarounds for AirPods stem detection during active microphone use.

Try it

I want to build a hands-free voice control interface for Claude Code on my local machine. 1. **Setup**: Install Parakeet v3 TDT for STT and Pocket TTS for TTS locally. 2. **Integration**: Create a custom MCP server to connect these models to Claude Code. Implement logic to detect AirPods stem clicks for interrupting narration and starting listening. 3. **Configuration**: Set up AirPlay streaming to a secondary display for visual monitoring. 4. **Dry Run**: Execute a simple debugging task on a non-critical script. Verify that voice commands are transcribed accurately and that TTS narration is clear. 5. **Human Check**: Confirm the agent's actions match my verbal instructions before proceeding to complex tasks. [Replaceable Inputs]: - Target hardware: [e.g., M4 Pro, AirPods Pro] - Task to debug: [e.g., a Python script for file organization] - Output location: [e.g., local terminal, TV via AirPlay] Measure success by the ability to complete the task without touching the keyboard. Handle errors by reverting to manual input if voice recognition fails twice consecutively.

More cases · Software & development

• Update a code base with Copilot chat and agent mode
• Check code against user intent with review agents and tests
• Build a budgeting app from a single prompt
• Build a mental models framework with a multi-agent dev team
• Build a instant-opening markdown viewer for macOS