Speech MCP Server

Local audio transcription via OpenAI Whisper, exposed as an MCP tool. Supports mp3, mp4, m4a, wav, ogg, flac, and webm formats.

Install


ark install marketplace/mcps/speech-mcp-server

Or with Helm:


cd mcps/speech-mcp-server
helm install speech-mcp-server ./chart -n default --create-namespace

Tool	Description
`transcribe_audio`	Transcribe an audio file using Whisper

Variable	Description	Default
`PORT`	Server port	`8080`
`BASE_DATA_DIR`	Base directory for audio files	`/data`
`WHISPER_MODEL`	Whisper model size (tiny, base, small, medium, large)	`small`
`CACHE_DIR`	Transcription cache directory	`/data/whisper-cache`

Used by the COBOL modernization bundle for transcribing interview recordings.