Model Context Protocol (MCP) • Native Agent Bridges

Bridge Your Voice Directly into AI Agents.

Speakeasy embeds a local Model Context Protocol (MCP) server, allowing Claude Desktop, Claude Code, and autonomous agent loops to query your meeting notes, search transcripts, and ingest spoken context directly.

1. Connect via Model Context Protocol (MCP)

Speakeasy ships with a local MCP stdio server. Add it to your claude_desktop_config.json or Claude Code MCP configuration:

~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "dictation-app": {
      "command": "/Applications/DictationApp.app/Contents/MacOS/dictation-cli",
      "args": ["mcp"]
    }
  }
}

2. Available MCP Tools Exposed to Agents

Once connected, Claude Code, Cursor, or any agentic system can invoke these local tools without user intervention:

MCP Tool Name Parameters Description
get_recent_dictations limit: number Retrieves your recent spoken thoughts and prompt history.
get_meeting_notes meeting_id: string Returns structured Markdown meeting minutes, speaker labels, and action items.
search_voice_history query: string Full-text search across all locally stored audio transcripts.
get_custom_dictionary none Returns your active technical vocabulary and acronym replacements.

3. CLI Piping for Unix Power Users

Use the built-in CLI to pipe audio or text directly through the Core ML speech engine:

# Transcribe a local WAV file to Markdown:

dictation-cli transcribe meeting-audio.wav > meeting.md

# Export today's action items as JSON:

dictation-cli meetings export --today --format=json | jq .action_items