Use a voice you prefer
Choose TalkToCursor Cloud, bring an ElevenLabs key, or run Voicebox locally.
One MCP server, your choice of voice
Make Codex sound like yours.
Add custom, managed, or fully local speech to Codex. Hear concise progress and completion updates without replacing Codex or its native Voice experience.
codex.response / speak
CodexTalkToCursorVoice
WHY ADD IT
Native Voice for conversation. TalkToCursor for control.
Codex already offers an integrated voice experience on supported surfaces. TalkToCursor fills the customization gap instead of trying to recreate it.
Get ambient task updates
Ask Codex to announce starts, milestones, questions, and completions through the MCP tool.
Keep alternative input options
Use Codex Voice, or dictate with Wispr Flow or Handy independently. TalkToCursor’s automatic loop remains Cursor-focused.
CODEX SETUP
Connect the existing package.
No Codex-specific fork is required. Codex launches the same TalkToCursor stdio MCP package used by Cursor.
Add TalkToCursor to Codex
Use the Codex CLI to write the MCP entry:
codex mcp add talktocursor -- npx -y talktocursorPrefer editing the file directly? Add this block to ~/.codex/config.toml.
[mcp_servers.talktocursor]
command = "npx"
args = ["-y", "talktocursor"]
startup_timeout_sec = 30Choose your speech provider
Open the shared settings app, then select TalkToCursor Cloud, ElevenLabs, or Voicebox.
npx -y --package talktocursor talktocursor-settingsTell Codex when to speak
Add concise voice behavior to your global ~/.codex/AGENTS.md or the project’s AGENTS.md.
## Voice feedback
Use the `speak` tool at task start, at important
milestones, when asking a question, and at completion.
Keep spoken updates concise. Do not read code, paths,
logs, or stack traces aloud.Confirm the connection
Restart Codex, inspect the registered MCP servers, then ask for a short spoken test.
codex mcp list
# In a Codex session:
Use the speak tool to say: "Codex voice is connected."
THE RIGHT TOOL
Use both for what each does best.
| Need | Codex Voice | TalkToCursor |
|---|---|---|
| Natural live conversation | Built in on supported Codex surfaces | Not a replacement for full-duplex Voice |
| Custom voice providers | OpenAI voice stack | Cloud, ElevenLabs, or local Voicebox |
| Automatic spoken milestones | Voice-session dependent | Rule-driven MCP calls |
| Alternative dictation | Native Codex input | Use Wispr Flow or Handy independently; the automatic loop is not yet verified for Codex |
DETAILS
Before you connect.
Does this replace Codex Voice?
No. It adds selectable text-to-speech and agent-initiated spoken updates. Use native Codex Voice when you want OpenAI’s integrated live conversation.
Why set a startup timeout?
The first npx launch may need time to download the package. A 30-second startup timeout prevents Codex’s shorter default from ending that first launch too early.
Does TalkToCursor Cloud work with Codex?
Yes. Cloud pairing and voice settings belong to the TalkToCursor package, so they are shared across supported MCP hosts on the same computer.
Give Codex the voice you want to hear.
Install the open-source MCP package, then choose local, bring-your-own-key, or managed speech.