[FEATURE]: Voice input MCP server for CLI terminal tools

Feature request

• I have searched for existing issues and confirmed this is not a duplicate

Problem

Terminal-based AI coding tools (OpenCode CLI, Claude Code, Cursor) cannot use voice input because terminal apps do not support clickable microphone icons. Users must type all prompts manually.

Proposed Solution

Add voice input support via a standalone MCP server (@opencode-ai/voice-mcp) that provides:

1. voice_transcribe — Record audio from microphone and transcribe to text
2. voice_type — Record, transcribe, and type at cursor position
3. voice_status — Check voice input availability

Implementation

• MCP server using @modelcontextprotocol/sdk
• Local Whisper transcription via faster-whisper (Python)
• Cross-platform recording (sox/ffmpeg)
• 100% local — no cloud API required
• Works with OpenCode CLI, Claude Code, Cursor, and any MCP-compatible tool

Usage

Users add to their config:

{
  "mcpServers": {
    "voice": {
      "command": "npx",
      "args": ["-y", "@opencode-ai/voice-mcp"]
    }
  }
}

Differentiation from existing issues

Existing issues (#41364, #33300, #35219, #37742, #39688) focus on desktop app voice input (GUI microphone button). This issue is about CLI/terminal voice input via MCP server, which works across all MCP-compatible tools.

Benefits

• Works in any terminal app (not just desktop)
• No subscription required
• 100% local processing
• Simple one-line config
• Cross-platform (macOS, Linux, Windows)

Testing

• Tested on macOS with faster-whisper base model
• Recording works with sox
• MCP server verified with tool listing

Next Steps

1. Review this issue
2. If approved, submit PR with implementation
3. Publish to npm as @opencode-ai/voice-mcp