MCP server

voxd is a Model Context Protocol server. AI agents and assistants can use it to speak, transcribe and polish audio on your computer.

Streamable HTTP: POST http://127.0.0.1:4870/mcp, authenticated with Authorization: Bearer vox_sk_…. Replies are plain JSON; there is no SSE stream, so GET /mcp returns 405. Batches are supported.
stdio: run python3 <bridge_path> with VOX_API_KEY set. bridge_path comes from GET /v1/connection. The bridge uses only the standard library. It finds voxd through voxd.json, or through VOXD_URL if set.

Protocol versions: 2025-06-18, 2025-03-26 and 2024-11-05.

Tools

Tool	Arguments	Result
`speak`	`text`, `voice?`, `speed?`, `save_to?`, `play?`	Take id, duration and audio path. `play: true` also plays it through the speakers.
`list_voices`	`language?` (prefix like `en`), `engine?`	Tab-separated `id, name, language, engine`, natural voices first
`transcribe`	`path`, `language?`, `timestamps?`	Text, or timestamped lines
`clean_audio`	`path`, `trim_pauses?`, `loudness?` (LUFS), `save_to?`	Before/after levels and the path of the cleaned audio
`convert_voice`	`path`, `voice` (`default` or `cv_…`), `save_to?`	Path of the converted audio (needs Chatterbox)
`add_pronunciation`	`term`, `say`	Adds or updates a pronunciation

save_to takes an absolute path ending in .wav, .mp3, .m4a/.aac, .opus/.ogg or .flac. The saved file is added to the export history.
Tool failures come back as results with isError: true and a readable message, so the agent can recover.

Example

curl http://127.0.0.1:4870/mcp -H "Authorization: Bearer $VOX_API_KEY" -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"speak","arguments":{"text":"Deploy finished","voice":"nova","play":true}}}'

The Integrations page in the app has setup instructions for Claude Code, Claude Desktop, Cursor, VS Code and more, with your port and key already filled in.