Speaks Claude's responses aloud using text-to-speech. This skill is always active - the speak tool must be the LAST thing called in every response.
View on GitHubvoice/skills/voice/SKILL.md
February 1, 2026
Select agents to install to:
npx add-skill https://github.com/nathanphelps/claude-voice-mcp/blob/main/voice/skills/voice/SKILL.md -a claude-code --skill voice-responseInstallation paths:
.claude/skills/voice-response/# Voice Response You MUST call the `speak` MCP tool as the **very last action** in every response. No text or tool calls should come after `speak`. ## Rules 1. **Speak last, always**: The `speak` tool call must be the final thing in your response. Write all your text and make all other tool calls first, then call `speak` as the last action. NEVER write text after calling `speak`. 2. **Every response**: Every response that contains text output to the user must end with a `speak` call. No exceptions. 3. **Be conversational**: Convert your written response into natural spoken language. Don't read markdown formatting, code blocks, file paths, or technical syntax aloud. Summarize those parts conversationally. 4. **Keep it concise**: For long technical responses, speak a brief summary (2-4 sentences) rather than the entire response. The user can read the details on screen. 5. **Skip for tool-only turns**: If your response is purely tool calls with no text output to the user, you don't need to speak. 6. **Don't announce the tool**: Just call speak naturally. Don't say things like "Let me read that aloud" in your written response.
Issues Found: