Voice for Agents (MCP)
Connect Spokenly to Claude Code, Codex, and Cursor so they can ask you questions by voice.
MCP (Model Context Protocol) integration lets AI coding tools talk to you through Spokenly. Spokenly has built-in setup for Claude Code, Codex, and Cursor, but works with any tool that supports MCP. When an agent needs clarification, it calls Spokenly's voice dictation tool and you answer by speaking instead of typing.
How It Works
- Spokenly runs a local MCP server at
http://localhost:51089. - Your AI tool connects to it and gets access to the
ask_user_dictationtool, which acceptsquestions: string[]. - When the agent needs input, it calls the tool and Spokenly starts recording.
- You speak your answer and press Enter. The transcribed text goes back to the agent.
This is faster than typing, especially for detailed explanations and code review feedback.
Requirements
- Spokenly 2.18.0+ (macOS sideload version, MCP is not available in the App Store build)
- One of the supported AI tools: Claude Code, Codex, or Cursor
MCP integrations require the sideload version of Spokenly. Download it from spokenly.app.
Setup
Open Spokenly and click on the integration you want to set up in the sidebar. Each integration shows a setup instruction block. Click Copy, paste it into Terminal, and restart your tool.
You can also set it up manually:
Testing
After setup, restart your AI tool and type this to verify the integration works:
Test Spokenly MCP by asking me 3 questionsThe agent should call Spokenly's voice dictation tool and you should see a recording prompt appear in Spokenly.
Speak Questions Aloud
Spokenly can read the agent's questions out loud using text-to-speech before starting the recording. This is disabled by default. To enable it, go to General Preferences > Voice for AI Agents (MCP) and turn on Speak AI questions aloud.
This feature uses cloud TTS and consumes credits. It may not be available on the free tier.