● OPEN SOURCE · SPEECH TO TEXT

Own the recorder.
Choose the intelligence.

Your audio. Your model. One transcript.

Keep the recorder you like. Send authorized audio to Microsoft, ElevenLabs or Groq, and get a consistent transcript your MCP client can use anywhere.

Get started ↗ Explore the providers →

Stable v1.2.2 / Apache 2.0 / Self-hosted

THE TRANSCRIPTION PIPELINEMCP

01Any authorized audioHTTPS

↓

02OpenTranscribeCheck capabilities. Route audio.↗

↓

MicrosoftMAIElevenLabsScribeGroqWhisper

↓

03One normalized transcript

SPEAKER_01

SPEAKER_02

Speaker turns when your model supports them.

↔Provider independentSame contract. Different intelligence.

◈Explicit capabilitiesRequire speaker turns. Never lose them silently.

◎Retention is your choiceOptional storage. Clear provider policies.

START HERE

Choose your setup

Use it locally, host it for your clients, or start with a recorder recipe.

01 / CONFIGURE

Connect your providers

Choose Groq, ElevenLabs, Microsoft, or a combination. Understand defaults, authentication and retention.

Configuration guide ↗ 02 / LOCAL

Start on Linux

Install OpenTranscribe Setup, add your keys and connect a local MCP client. No Python setup needed.

Linux desktop guide ↗ 03 / HOSTED

Deploy on Scaleway

Run a secure MCP endpoint with scale-to-zero. Follow the Terraform setup, then upgrade with verified images.

Deployment guide ↗

Working with a Plaud recorder? Start with the Ubuntu walkthrough. Prefer Python or Docker? Use the repository quickstart.

A SMALL, FOCUSED TOOL

Audio in. A portable transcript out.

OpenTranscribe handles transcription. Summaries, storage libraries and downstream workflows belong to your MCP client. Use only audio you are authorized to process; speaker labels never identify real people.

Architecture →

Security and privacy →

Publication protocol →