Voice Input for Claude Code and Cursor AI

We wrote previously about why voice dictation and AI coding are a natural fit. The short version: AI coding tools run on natural-language prompts, and you can speak natural language roughly twice as fast as you can type it. Voice input makes vibe coding faster and produces better prompts because you naturally include more detail when speaking.

This post is the practical follow-up. Below are exact, step-by-step instructions for setting up voice input in Claude Code and Cursor AI using Voiced, a local dictation app for Mac that works in any text field.

What you need

Before starting, make sure you have:

• Voiced installed. Download Voiced, drag it to Applications, grant microphone and accessibility permissions, and wait about a minute for the speech model to download on first launch. No account or API key required.
• Claude Code or Cursor (or both) installed and ready to use.
• A working microphone. Your MacBook's built-in mic works fine. An external mic will give slightly better accuracy in noisy rooms, but it's not necessary.

Once Voiced is running, you'll see a small icon in your menu bar. The default hotkey is the right Option key — hold it to record, release to transcribe. You can change this in Voiced's preferences to any key you like.

Voice input for Claude Code (step by step)

Claude Code is a terminal-based AI coding tool from Anthropic. You run it in your terminal and interact entirely through text prompts. That makes it a perfect candidate for voice dictation — it's all text input, no buttons or menus to click.

1. Open your terminal and start Claude Code. Run claude in your project directory as usual. You should see the prompt waiting for input.
2. Hold your Voiced hotkey. By default that's the right Option key. You'll see a small indicator confirming Voiced is listening.
3. Speak your prompt naturally. For example: "Refactor the authentication middleware to use JWT tokens instead of session cookies. Make sure to update the tests and add error handling for expired tokens."
4. Release the hotkey. Voiced transcribes your speech in under a second and pastes the clean text directly into the terminal at your cursor position.
5. Review the text and hit Enter. Claude Code takes it from there.

That's it. No copy-paste, no switching windows. You speak directly into the terminal prompt and Claude Code receives your full, detailed instruction. Since Claude Code often requires multi-sentence prompts with specific context about your codebase, voice input saves a meaningful amount of time on every interaction — a difference that adds up quickly over a full day of coding.

Voice dictation for Cursor AI (step by step)

Cursor is a code editor with AI built in. It has multiple surfaces where you type prompts: the chat panel, inline edit (Cmd+K), and the composer. Voiced works with all of them.

Chat panel

1. Open the Cursor chat panel (Cmd+L or click the chat icon).
2. Click into the chat input field so your cursor is active there.
3. Hold your Voiced hotkey and speak. Describe your question or request: "Why is the useEffect in the Dashboard component running twice on mount? I only want it to fetch data once."
4. Release the hotkey. Your transcribed prompt appears in the chat field.
5. Hit Enter to send it to the model.

Inline edit (Cmd+K)

1. Select code in the editor and press Cmd+K to open the inline edit prompt.
2. Hold your hotkey and describe the change: "Add input validation for the email field. Check that it's a valid email format and show an error message below the input if it's not."
3. Release and hit Enter. Cursor applies the edit inline.

Composer

1. Open the composer (Cmd+I).
2. Hold your hotkey and speak your multi-file instruction: "Create a new API route at /api/users that supports GET for listing all users and POST for creating a new user. Add the corresponding database queries in the models folder and a test file."
3. Release and submit.

The process is the same in every case: click into the text field, hold and speak, release. Voiced doesn't care which field it is — it pastes text wherever your cursor is.

Tips for effective voice prompts

Include context. AI coding tools produce better results when you explain the why, not just the what. Instead of "add a loading spinner," say "add a loading spinner to the user list page because the API call takes two to three seconds and right now it just shows a blank screen." Voice makes this easy because you don't have to type out the extra sentences.

Speak in full sentences. When typing, developers tend to abbreviate. When speaking, let yourself use complete thoughts. The AI model benefits from clear, unabbreviated instructions.

Don't worry about filler words. Voiced's Smart Cleanup feature automatically strips "um," "uh," "you know," "like," and other verbal tics from your transcription. Speak naturally and the output reads clean. This matters more than you'd think — a prompt full of filler words can confuse the model, and Smart Cleanup eliminates that problem before it starts.

Use it for follow-up prompts too. Voice input isn't just for the initial request. When the AI generates code and you need to give feedback — "that's close, but move the error handling into a try-catch block and use a custom error class instead of a generic Error" — speaking your iteration is faster than typing it.

Why local processing matters for developers

Your coding prompts contain information about your codebase: architecture decisions, business logic, API structures, database schemas. When you dictate these prompts, that information is in the audio.

Cloud-based dictation tools send your audio to external servers for processing. For developers working on proprietary code, that's a problem. You might work behind an NDA. Your company might have policies about where data can be sent. Even without formal restrictions, sending a stream of audio containing detailed descriptions of your codebase to a third party is a needless risk.

Voiced processes everything locally on your Mac using an on-device speech model. Your audio never leaves the machine. No cloud calls, no accounts, no telemetry. For a deeper look at how Voiced compares to cloud-based alternatives, see our comparison of the best voice-to-text apps for Mac.

Getting started

Setting up voice input for Claude Code or Cursor takes about 60 seconds:

1. Download Voiced and drag it to Applications.
2. Grant microphone and accessibility permissions when prompted.
3. Wait for the on-device model to download (about a minute on first launch).
4. Open Claude Code or Cursor, hold your hotkey, and start talking.

Voiced is a one-time purchase with a free 10-day trial. No subscription, no cloud dependency, no API keys to manage. Just a voice-to-text app for Mac that works everywhere you type — including the AI coding tools where it makes the biggest difference.