Transcribe audio from Claude or Cursor via MCP

Your AI can already transcribe. With the Vook.ai MCP, it does it better: up to 99% accuracy, speakers identified, no training on your audio, hosted in France. Send a recording, then read, summarize and export the transcript without leaving the conversation.

Get startedSet-up guide
Claude Desktop · conversation

Transcribe interview.m4a with speakers and give me the three main points

vook · transcribe_file · interview.m4a · 23:14

Done. Two speakers, 23 minutes. The three main points:

  1. The launch moves to March, pending the beta results.
  2. Pricing stays monthly; the annual plan is parked.
  3. Next step: a written proposal by Friday.

Transcript, summary and exports stay in the conversation

Up and running in three steps

1

Create your key

In your Vook.ai account, under API. It stays on your machine.

2

Add vook-mcp to your AI

One command in Claude Code, or a few lines in the MCP settings of Claude Desktop and Cursor.

3

Ask for a transcript

Name a file or a URL in the conversation. Your AI does the rest.

Claude Code · one line
claude mcp add vook \
  --env VOOK_API_KEY=vk_live_… \
  -- npx -y vook-mcp

# Claude Desktop, Cursor: add "npx -y vook-mcp" to your MCP config
Set-up guide for Claude Desktop, Cursor and more

What your AI can do with the Vook.ai MCP

Four things to ask for. Your AI picks the right tool each time.

Transcribe

A local file up to 6 GB or a URL, with speaker identification.

Read

The transcript, page by page, so long recordings fit the conversation.

Summarize

A detailed summary in the language of the audio.

Export

SRT, Markdown and HTML in the conversation; PDF and Word written to disk.

Ask it like this

Transcribe interview.m4a with speakers and give me the three main points

Make subtitles for this video URL and save them as SRT

List my transcriptions from this week and summarize the longest one

A transcription API isolated from any AI model training

Hosted on European infrastructure with AES-256 encryption and full GDPR compliance.

EU infrastructure

All servers, processing pipelines, and storage operate within European data centers, fully compliant with GDPR.

AES-256 encryption

Data is encrypted at rest across all storage layers, from upload to retrieval.

Zero model training

Audio files and output transcripts are never used to train internal or third-party AI models.

You control deletion

Transcripts and audio stay until you delete them, from the app or with one API call. Working audio used during processing is deleted after a day.

Need legal documentation for your compliance team?

Access our DPA →

Transparent pricing

Fixed rates per audio hour, no hidden platform fees or add-on costs. Try it for free with one transcription a day, pay only when you need more.

STANDARD
$0.60/ hour of audio
Pay only for successful jobs.
Silent or corrupted files are never billed.
Sign up →
HIGH VOLUME
From $0.20/ hour of audio
Volume discounts, à la carte features, and custom features built for your use case. Pay only for what you enable.
Talk to sales

Give your AI a transcription tool

Create your key, add vook-mcp, and ask for your first transcript.

Get startedSet-up guide

MCP transcription: frequently asked questions

Can't find your question? Send us a message!

What is MCP transcription?

MCP transcription means your AI transcribes audio through an MCP tool. MCP (Model Context Protocol) is the standard that lets an AI call outside tools. The Vook.ai MCP is a small program your AI starts on your machine. It turns a request like "transcribe this file" into calls to Vook.ai with your key, and hands the result back to the conversation.

Which AI apps are supported?

Claude Code with one command, Claude Desktop and Cursor through their MCP configuration. Any other MCP client that runs local servers works with the same command. Node.js 22 or later is required on the machine.

Does it work with claude.ai or ChatGPT?

Not today. The Vook.ai MCP runs on your machine, inside your desktop AI app. claude.ai and ChatGPT only connect to hosted MCPs, which we have not released.

Is Vook.ai's MCP transcription free?

You get 1 free transcription a day with your API key. Paid transcription costs $0.60 per hour of audio transcribed, from prepaid credit, with no subscription. Listing, reading and exporting a transcript cost nothing.

How accurate is MCP transcription?

Up to 99% accuracy on clear audio, with at least 90% guaranteed across all transcriptions. Speakers are identified and every segment carries its timestamps.

Can my AI delete my transcripts?

No. The MCP has no delete tool. An AI reads transcript text as input, and a recording could contain instructions, so nothing it hears can remove your data. You delete from the Vook.ai app.

Which files can it transcribe?

The same formats as the app: audio (MP3, WAV, M4A, FLAC, OGG, OPUS, AAC) and video (MP4, MOV, MKV, WEBM), up to 6 GB per file and 5 hours of audio. Any other extension is refused before anything is sent.

Which export formats does the MCP give?

SRT, Markdown and HTML, returned in the chat. PDF and DOCX are written to a file on your machine.

Where is my audio stored, and for how long?

In France (EU), encrypted with AES-256 at rest, and never used to train models. Your audio and transcripts stay in your account until you delete them in the Vook.ai app.