Install
claude plugin marketplace add mbailey/voicemode
claude plugin install voicemode@voicemodeWho it is for
You already use Claude Code and want to keep the conversation going with natural speech when your hands are busy and typing is inconvenient. Official text says voice is not meant to replace typing; it keeps the conversation usable when typing is impossible. For people who accept a real voice in-and-out, not just text echo.
Platform excerpts cover Linux, macOS, Windows (native or WSL), and NixOS, Python 3.10-3.14. Current notes do not list a required microphone model, speaker requirements, or whether a cloud speech service is mandatory. If you cannot install a plugin for Claude Code, or cannot allow voice-related tools, this guide cannot help you.
First try in 3 minutes
Prerequisites: Claude Code is available, Python is 3.10-3.14, and the current environment can speak and listen. The official recommended path is add the marketplace first, then install the plugin. Do not start with a bypass MCP command:
claude plugin marketplace add mbailey/voicemode
claude plugin install voicemode@voicemode
After install, restart or reopen Claude Code once and confirm the plugin is present. For the first try, use only mcp__voicemode__converse as named in the excerpt. This is not a read-only query: it enters a voice conversation, may use local audio devices, and will hand your spoken words to the agent. Before you start, confirm you are in a place where you can speak, can listen, and do not mind this conversation being processed by the model, then make the call.
Observable success: the conversation continues in voice form, not that the plugin install command itself returned success. Current notes do not give a fixed test phrase or a log keyword. If the client shows a permission prompt, read that it is allowing mcp__voicemode__converse before you agree. Do not call mcp__voicemode__service on the first try — the excerpt only lists it on the permission list and does not say what it does.
The materials also have claude mcp add --scope user voicemode -- uvx --refresh --from voice-mode voicemode-mcp-launcher, which is another install path. The first try should use only the plugin marketplace above; do not configure two entry points at once.
Use cases
Case 1: when typing nearby is a poor fit, use mcp__voicemode__converse to turn an in-progress Claude Code task into a voice round-trip. Current notes only promise "natural voice conversation" and do not list language, barge-in, noise reduction, or parameters that turn speech into an editable draft. Success is that you can understand the reply and continue the next turn, not that a particular text log appears.
Case 2: when voice capability must be allowed explicitly, the permission list includes both mcp__voicemode__converse and mcp__voicemode__service. The excerpt does not explain what service does, whether it changes local services, or how to fall back on failure. This case therefore stops at "allow converse from the list; service stays unexplained." Do not invent steps to restart a service or install a system daemon.
Limits
The official page is
https://github.com/mbailey/voicemode
Do not read this as "no network, no local reads, no privacy boundary": a voice conversation by default captures what you say. Python must be 3.10-3.14. The platform list stops at Linux, macOS, Windows (native or WSL), and NixOS. The excerpt has no uninstall, microphone-off, or conversation-revoke steps. The service tool has no usage beyond its permission name. A bypass uvx launch command exists, but do not let it compete with the plugin marketplace as a second main path. Quotas, billing, and required cloud keys are not written.