The original integration, with Cursor rules plus optional Wispr Flow or Handy auto-submit on macOS.
OPEN CURSOR GUIDE →One speech layer for your coding agents.
TalkToCursor gives compatible AI coding hosts a shared speak tool and selectable voices, with host-dependent dictation options. Keep the assistant you prefer and choose how it sounds.
Same MCP package. Different connection steps.
Use the guide for your host. Each one includes the right configuration format, persistent instruction file, and host-specific caveats.
Spoken output that complements native dictation, configured with Claude Code MCP and CLAUDE.md.
Custom and local TTS that complements Codex Voice, configured through Codex MCP and AGENTS.md.
Spoken agent output that complements Antigravity’s native live transcription, configured through MCP and global or workspace Rules.
OPEN ANTIGRAVITY GUIDE →Can another IDE or agent host use it?
- It must support local stdio MCP servers.
- It must be able to launch Node.js 18+ commands such as
npx. - It must let the model call a tool named
speak. - For automatic narration, it should support persistent agent instructions or rules.
Start with the host’s local MCP configuration.
MCP clients use different file locations and formats. Add the equivalent of this JSON entry, restart the host, and confirm that the speak tool appears.
{
"mcpServers": {
"talktocursor": {
"command": "npx",
"args": ["-y", "talktocursor"]
}
}
}
npx -y --package talktocursor talktocursor-settings
Use the `speak` tool for concise task starts,
important milestones, questions, and completions.
Do not read code, paths, logs, or stack traces aloud.
Portable settings, three provider paths.
Managed Google speech, portable settings, and no provider key stored locally. Explore Cloud
Bring your own key for a broad voice library and detailed controls. Browse ElevenLabs
Run speech and voice profiles locally for a private, free option. View Voicebox
Compatibility depends on the host’s MCP implementation. Hosts not listed above are not claimed as verified until their configuration and tool behavior have been checked.
Keep your coding agent. Choose its voice.
Install the open-source MCP package once and connect it wherever compatible.