How to talk to Claude Code with your voice

You type the same long paragraph to your coding agent twenty times a day. You could say it in a third of the time.

Short answer: The fastest way to talk to Claude Code is its built-in /voice command, which turns on push-to-talk in the terminal but streams your audio to Anthropic, needs a claude.ai login, and does not work over SSH. For voice that types into any terminal or editor, including an SSH session, use a dictation tool that pastes at your cursor, such as MightyMouse, which transcribes on your own Mac for free.

Download free for Mac

Voice is a better fit for coding agents than for almost anything else you dictate. When you write code, you need exact symbols: brackets, camelCase, a semicolon in the right place. When you prompt an agent, you write prose. "Fix the login redirect so it keeps the query string, and add a test for it." That is a sentence, and a sentence is what speech is good at.

So the real question is not whether to talk to Claude Code. It is which way to do it. There are five, and each one gives something up. Here they are, with what Anthropic and the vendors say on their own pages.

Option 1: the built-in /voice command

Claude Code has voice input of its own. Per Anthropic's voice dictation docs (as of October 2026), you run /voice, then either hold Space while you speak or switch to tap mode with /voice tap. Your words appear in the prompt as you talk, so you can mix typing and speaking in one message. The docs say transcription is tuned for coding words like regex, OAuth and JSON, and that your project name and git branch are added as hints.

It is good, and if you only ever use Claude Code in a local terminal, it may be all you need. The same docs are plain about the catch, though:

• Audio is streamed to Anthropic's servers for transcription. In their words, "Audio is not processed locally."
• It needs a claude.ai account. It is not available if Claude Code runs on an Anthropic API key, Amazon Bedrock, Google Cloud, or Microsoft Foundry.
• It needs a local microphone and, per the docs, "does not work in cloud sessions or SSH sessions."
• It lives inside the Claude Code CLI and the VS Code extension. It does not help you in Codex, Cursor's chat box, or a browser.

If any of those is a dealbreaker, read on. Anthropic also says transcription does not use your Claude messages or count against your usage limits, which is a real plus.

Option 2: macOS Dictation

Apple's Dictation works in any text field. You turn it on in System Settings, under Keyboard, then Dictation, and you pick a shortcut to start it. Apple's support page says you can dictate text of any length with no timeout, and that it stops after 30 seconds of silence. It can also keep general dictation on your device instead of sending it to Siri servers, which you can check in Keyboard settings.

It is free and already installed. The catch for agent work is that it is built for typing into one text field, not for pasting each sentence as you finish it. If it goes quiet in your terminal, here is why Mac dictation stops working and what to check.

Option 3: Talon, if you want to code entirely by voice

Talon is the serious tool for people who need or want to control the whole computer by voice: writing, coding and browsing, with custom commands you define in files. Its site lists macOS, Windows and Linux downloads and an optional paid Talon+ tier.

If you have a repetitive strain injury or you want to type symbols by speaking them, Talon is the answer and nothing else on this list comes close. It is also a real commitment. You learn a command vocabulary, and you set it up to fit you. That is more than most people want just to give a coding agent a paragraph of instructions.

Option 4: cloud dictation apps

Apps that send your audio to a server and return polished text do the best cleanup. They remove "um," fix punctuation, and sometimes rewrite for tone. If you like that, the best dictation apps for Mac compares them. The tradeoffs are a subscription, your voice leaving your computer, and whatever each app does when the cursor is in a terminal or an SSH session.

Option 5: local Whisper that types at your cursor

MightyMouse is a Mac app that turns your voice into text on your own computer, using whisper.cpp, or NVIDIA's Parakeet model for the languages it supports, both running locally. Your audio is not uploaded, and transcription is free. What you say gets pasted wherever your cursor is, so it works the same in Terminal, iTerm, VS Code, Cursor, a browser box, or the Codex CLI. Claude Code is just a text box that happens to live in a terminal.

The mode that matters for agents is hands-free, and it runs from the iPhone you have paired to your Mac as its mic. Here is how it works:

1. Click into the terminal or editor where you talk to your agent.
2. On your phone, press and hold the mic button for about a second without talking. A pill says hands-free is on.
3. Talk. When you pause, your Mac transcribes that stretch of speech and pastes it at the cursor. Then it keeps listening.
4. Say the next thought, pause, and it pastes again. One utterance, one paste.
5. Tap the mic once to stop. Then read what landed and press Enter yourself.

Because each pause becomes its own paste, long prompts build up in pieces while your hands stay free. You can pace around the room explaining a refactor, and the paragraph is waiting when you sit down. Plain dictation only pastes, so pressing Enter stays your job, which is exactly what you want when the thing on the other end can edit your files.

Why this works over SSH

This is where the "pastes at the cursor" approach earns its keep. MightyMouse pastes into whichever Mac window is focused by sending a real Cmd+V keystroke. If that window is a terminal with an SSH session running Claude Code on a remote box, the terminal just receives pasted text and forwards it, the same as if you had typed it. Nothing has to be installed on the server, and the server needs no microphone. Anthropic's /voice cannot do this, because its docs say it needs a local microphone and does not work in SSH sessions. For the deeper why, see why dictation breaks over SSH.

Talk from across the room with your iPhone

The same dictation also works from your phone. In the page MightyMouse serves to your iPhone's Safari, you tap the mic and talk. The phone only listens. The transcription still runs on your Mac, and the result is pasted at your Mac's cursor, as if you had dictated at the desk. That is handy when you are sitting back from the keyboard, standing at a whiteboard, or walking the dog while the agent works. The phone as a microphone is covered in using your iPhone as a mic for your Mac.

If you want to check on an agent or answer its questions when you are fully away from the desk, that is a different job. Claude Code on your iPhone and running coding agents from your phone cover it.

How to prompt by voice without regrets

Speaking a prompt is a different skill from typing one. These habits make the difference between a clean result and a mess you have to fix.

• Say file names slowly and fully. "In src slash auth slash login dot ts" lands better than "the login file." If a path is awkward to say, type just that part.
• Give the task a shape. Say "Three things. One, ... Two, ... Three, ..." Models follow a spoken list as well as a typed one, and it stops you rambling.
• State the finish line. End with what done looks like: "Run the tests and stop when they pass." Speech makes it easy to forget the end condition.
• Name what not to touch. "Do not change the public API" is easy to say and easy to skip when typing.
• Read before Enter. Look at the pasted text every time. A misheard word in a prompt can send an agent the wrong way, and it is cheaper to fix a sentence than a diff.
• Dictate the intent, type the identifiers. Talk through the plan. When you need useAuthSession or a regex, type it, or say a plain description and let the agent find the symbol.
• Pause between ideas. In hands-free mode a pause starts the paste, so a short breath splits a long thought into clean chunks.

What this costs you: the honest limits

• Code symbols by voice are hard. None of the tools in options 2, 4 or 5 will reliably turn "open paren, camel case foo bar" into code. Talon is the one that treats that seriously. For everything else, voice is for prose and type the symbols.
• Local cleanup is weaker. MightyMouse pastes what Whisper heard, with no rewriting pass. A cloud app that polishes your text will often read cleaner. You trade polish for privacy, price and a prompt that is exactly what you said.
• It is a Mac app. It runs on macOS. A Windows version is in beta, and it is not on Linux. If you code on Linux, /voice or Talon fit better.
• The Mac has to be awake. It cannot dictate into a sleeping Mac or past a locked screen. That applies to talking from your iPhone too.
• Each utterance has a short wait. Transcription runs after you pause, so text arrives a moment after you stop talking, not word by word as you speak. /voice shows words live as you talk, which some people prefer.
• Noise and accents matter. Any speech model gets things wrong in a loud room. Read before you press Enter.

Which one should you pick?

If you use Claude Code only in a local terminal, you have a claude.ai account and you are fine with cloud audio, try /voice first. It costs nothing to try. If you work over SSH, switch between Claude Code, Codex and Cursor, want your voice to stay on your computer, or want hands-free mode and a phone mic, use local Whisper through dictation that works in any Mac app. If you cannot use a keyboard, learn Talon. You can mix these. Plenty of people keep /voice for quick prompts and a hands-free tool for long ones.

For more on the hands-free mode, see hands-free dictation on a Mac. To see how local transcription is built, see running Whisper locally on a Mac.

Common questions

Can you talk to Claude Code with your voice?

Yes. Claude Code has a built-in /voice command that records while you hold Space, or in tap mode. Audio goes to Anthropic for transcription and you need a claude.ai account. Any dictation tool that pastes at your cursor also works in the terminal.

Does Claude Code voice mode work over SSH?

No. Anthropic's docs say voice dictation does not work in cloud sessions or SSH sessions, because it needs a local microphone. A dictation app on your Mac that pastes into the focused terminal window works, since the terminal forwards the pasted text to the remote session.

Is Claude Code voice dictation processed locally?

No. Anthropic's docs say /voice streams your recorded audio to Anthropic's servers and that audio is not processed locally. A local Whisper dictation app keeps the audio on your own computer instead.

Can I talk to Claude Code from my phone?

Yes, in two ways. You can use your iPhone as a microphone so your words are typed at your Mac's cursor, or you can follow and answer an agent from your phone. The Mac still runs Claude Code in both cases.

Is voice coding practical for writing actual code?

It is practical for prompting an agent, because prompts are sentences. Dictating exact code symbols is much harder. Tools like Talon are built for that, while ordinary dictation works best for the plan and not for the identifiers.

What is the best way to dictate into Claude Code on a Mac?

If you only use it in a local terminal and accept cloud audio, /voice is the simplest. If you want it to work in any terminal or editor, over SSH, with private on-device transcription, use a local dictation app that pastes at the cursor.

To talk to any terminal or editor and keep your voice on your own computer, try MightyMouse, free voice typing that runs on your Mac.

Read next

• Why dictation breaks over SSH
• Hands-free dictation on your Mac
• Claude Code on iPhone
• Use your iPhone as a microphone for your Mac