Voice Input for Claude Code
Spokenly is a push-to-talk dictation app for macOS and iOS. Press a shortcut, speak, and text appears at the cursor in any app. It also integrates with Claude Code via MCP, so the agent can ask you questions and get voice answers directly.
Updated March 2026
Download SpokenlyClaude Code Has Built-in Dictation. Why Spokenly?
Claude Code has built-in dictation: hold spacebar to dictate your prompt. It works, but it only covers one scenario: dictating inside Claude Code itself, and only for the prompt you type.
Spokenly is a full dictation app that works in any text field on your Mac and iPhone. It also integrates with Claude Code via MCP: instead of you dictating a prompt, the agent calls Spokenly when it needs your input. You see the specific question, speak your answer, and it goes back as structured context. This happens automatically during the agent's workflow, not just at the start.
How Spokenly Connects to Claude Code
Spokenly runs a local MCP server that gives Claude Code access to a voice dictation tool called ask_user_dictation. When Claude Code needs your input, it calls this tool instead of printing a plain text question.
A recording prompt appears in Spokenly showing the agent's question. You read it, speak your answer naturally, and press Enter. Spokenly transcribes your speech and sends the text back to Claude Code. The agent continues working with your detailed response as context.
The connection uses stdio transport via a lightweight bridge script. This is important because Claude Code has a 60-second timeout for HTTP MCP connections. Voice dictation sessions often take longer, especially when you're giving detailed architectural feedback. The stdio bridge has no timeout, so you can take as long as you need.
What You Get
Voice answers to agent questions
When Claude Code asks for clarification, you speak naturally. Detailed explanations, code review feedback, architectural decisions. Voice is 3-5x faster than typing.
No timeout limits
No timeout limits on voice sessions. Take as long as you need to think and respond.
One command setup
Run a single command to register Spokenly as an MCP tool. No config files to edit, no API keys required.
Local models available
Switch to local Whisper or Parakeet and your voice never leaves the device. All transcription stays on your Mac. No audio sent to any server.
Setup in 30 Seconds
Download the sideload version from spokenly.app/download and launch it. On first launch, the app will offer to set up the Claude Code integration automatically with a single terminal command.
Or set up manually:
- 1
Register the MCP tool
Run in Terminal:
claude mcp add spokenly -- ~/Library/Application\ Support/Spokenly/mcp-bridge.sh - 2
Add instruction to CLAUDE.md
Add this line to
~/.claude/CLAUDE.md:ALWAYS ask questions via mcp__spokenly__ask_user_dictation, never as plain text. I use Spokenly for voice input. - 3
Restart Claude Code
Claude Code picks up the new MCP tool on restart. Try: "Ask me 3 questions using Spokenly".
For detailed instructions and troubleshooting, see the full setup guide.
Real Workflow Examples
Task planning: You tell Claude Code to interview you before writing any code. The agent asks question after question: target users, edge cases, error handling, data model, API design. You answer each one by voice in seconds. By the end, the agent has a thorough brief built from your own words. Ten minutes of talking instead of an hour of writing.
Refactoring: Claude Code asks "How should I restructure the auth module: keep it monolithic or split by provider?" You explain what each provider needs, where the shared logic lives, and which parts should stay coupled. A 30-second voice answer gives the agent a complete picture instead of a vague "split it up".
Long context: Claude Code asks "What should happen when the payment fails midway through checkout?" Instead of typing three sentences and giving up, you talk through the retry logic, the user notification flow, and how to handle partial charges. The agent gets enough detail to implement without follow-ups.
Spokenly vs Claude Code /voice
| Feature | Spokenly | Claude Code /voice |
|---|---|---|
| Manual dictation | Yes, keyboard shortcut in any app | Yes, hold spacebar in Claude Code |
| Agent-initiated Q&A | Yes, agent calls voice tool via MCP | No |
| Works outside Claude Code | Yes, any app on your Mac | No |
| Local/offline models | Yes (Whisper, Parakeet) | No |
| Custom AI prompts | Yes | No |
| iOS app | Yes | No |
| Price | Free (local + own API keys) | Included with Claude Code |
You can use both together. /voice for quick prompts, Spokenly for agent-initiated questions and dictation everywhere else on your Mac.
Frequently Asked Questions
How do I add voice input to Claude Code?
Download Spokenly from spokenly.app/download, then run: claude mcp add spokenly -- ~/Library/Application\ Support/Spokenly/mcp-bridge.sh. Restart Claude Code and it will automatically use voice dictation when asking questions.
Is voice input for Claude Code free?
Yes. Spokenly's MCP server is included in the free tier. Local speech-to-text models (Whisper, Parakeet) are free with no account, no time limits, and no word caps. Cloud models are available via Pro plan or with your own API keys.
Does it work with Claude Code on all platforms?
Spokenly is a macOS app, so the MCP integration works with Claude Code running on Mac. The MCP server uses localhost, so both Spokenly and Claude Code need to run on the same machine.
Why use stdio transport instead of HTTP for Claude Code?
Claude Code has a 60-second timeout for HTTP MCP connections. Voice dictation can take longer than 60 seconds if you're giving a detailed answer. The stdio bridge script has no timeout limit, so your recording sessions can be as long as needed.
Can I use voice and keyboard input together?
When Claude Code calls the voice tool, a recording prompt appears. Speak and press Enter, or press Escape to skip and type instead. You always have the choice.
Does my voice data stay private?
With local models, all speech-to-text processing happens on your Mac. Audio never leaves your device. You can enable Local Only Mode in Spokenly to ensure no network requests are made.
What happens when Claude Code asks multiple questions?
Spokenly handles multi-question tool calls natively. Each question gets its own recording session, so you answer them one by one.
Voice Input for Other Tools
Talk to Claude Code
Quick setup. Free with local speech-to-text models.
Download Spokenly