Cadence Code

Fully local voice conversations for coding agents on Apple Silicon. A stdio MCP server owns the selected text-to-speech and speech-to-text models directly through MLX Audio, so speech never leaves the machine and no cloud speech service is used. Models are user-selectable. 8 tools: `voice_start`, `voice_speak`, `voice_listen`, `voice_interrupt`, `voice_stop`, `voice_models`, `voice_configure`, `voice_status`. Works with Claude Code, Codex, Cursor, and Antigravity. Install: `/plugin marketplace add michael-L-i/cadence-code`. Python, MIT.

Project

Browse this MCP server in the TensorBlock MCP Index.