Voibe Transcription
Give ears to your AI agent.
What it can do
- Create Transcription Job: Start a transcription job and get back the two ways to hand over the audio. Call this FIRST, before you have sent any audio. It returns: - `upload_url`: PUT the raw audio byt
- Get Transcript: Read a job: its status, and when it is finished the transcript and summary. Returns immediately, whatever the state. `status` is QUEUED (waiting for the audio, or just started), PROCES
- List Transcripts: List this account's transcription jobs, newest first. Use it to find a job whose id the person does not remember. Segments are never included here; call `get_transcript` for one job'
What data it sees
Do you need an account
No: the server works without sign-in
Give ears to your AI agent. Voibe is a remote MCP server that turns recordings into text your agent can work with — meetings, interviews, calls, lectures, podcasts, voice memos. Connect once, then just ask. Same server, whether you're in Claude, Cursor, Codex, ChatGPT, VS Code, Claude Code, Hermes or OpenClaw.
How it works
Ask your agent to transcribe a recording. The raw transcript comes back in the chat with speaker labels, timestamps and a summary — so the real work starts immediately.
- In a chat app (Claude, ChatGPT): the agent hands you an upload link. Drop the file in, go back to the chat.
- In a coding tool (Claude Code, Cursor, Codex): point at a file — or a whole folder of recordings in one instruction — and the agent reads it from disk and uploads it for you.
You can also ask it to list your recent transcriptions or check your remaining minutes.
Connect
Address: https://api.getvoibe.com/mcp
Add it as a custom/remote MCP connector, then sign in and approve — the app gets its own access that renews itself, so you approve once. New accounts get 15 free minutes, no card required. For headless tools (a server or script), use an API key in the Authorization: Bearer header instead.
Tools
| Tool | Does |
|---|---|
create_transcription_job |
Start a job. Options: speaker labels, summary prompt, webhook. Returns the upload link. |
get_transcript |
Status, then the summary and transcript text. Per-speaker lines with timestamps on request. |
list_transcripts |
Your jobs, newest first. |
get_balance |
Minutes remaining. |
An app cannot see or create your API keys, buy minutes, change billing, or sign in as you anywhere else. What's said in a recording can't change what the app does with your account.
What you can do with the transcript
- A meeting becomes minutes with decisions and action items.
- A user or hiring interview becomes attributed quotes.
- A sales call becomes the follow-up email.
- A podcast episode becomes show notes and chapter markers.
- A lecture becomes study notes.
- A voice memo becomes a to-do list.
- A support call becomes a score against your rubric.
- Several calls become one report — e.g. what customers said about pricing.
What comes back
- Who said what: every speaker told apart, every line with start and end times. When people say their names, the summary uses them.
- A summary in the shape you ask for: decisions only, action items with owners, bullet points.
- Multilingual: the recording doesn't have to be in English.
- Files up to 200 MB in common audio and video formats — mp3, wav, m4a, mp4, flac, ogg, webm and more.
Your audio
Audio is deleted the moment the transcript exists, and never used to train models. Transcripts and summaries stay readable for 24 hours — transcribe and work on the result in the same sitting.
Pricing
Sign in for an account with 15 free minutes — no card, no subscription. After that, buy one-time minute packs that never expire. You pay per second of audio, and only when a job finishes. Failed jobs cost nothing.
From the makers of the Voibe dictation app.
Server tool list (4)
Raw names from tools/list. Only developers need these.
| create_transcription_job | Start a transcription job and get back the two ways to hand over the audio. Call this FIRST, before you have sent any audio. It returns: - `upload_url`: PUT the raw audio bytes here with header `Content-Type: application/octet-stream`. Use this when you can read the file yourself. Transcription starts on its own the moment it lands. - `upload_page_url`: a web page the person opens to drop the file in from their browser. Give them this link when you cannot reach their file. Do not ask for the audio as text or base64, and never try to send file bytes as a tool argument. `diarize: true` labels who spoke (speaker separation; off by default — it is the slow part, switch it on only when speakers matter). `prompt` steers the summary only, never the transcript. `webhook_url` is optional and only useful to a caller running a server. Billing is per second of audio and only when a job finishes. Creating a job costs nothing, but it is refused with "no minutes left" when the balance is zero or below — then tell the person to buy more minutes. After the audio is sent, call `get_transcript` with the job id. Work takes a fraction of the audio's length, so wait a little before the first check and do not call it in a tight loop. |
| get_transcript | Read a job: its status, and when it is finished the transcript and summary. Returns immediately, whatever the state. `status` is QUEUED (waiting for the audio, or just started), PROCESSING, DONE or FAILED. When it is not finished yet, say so and check again in a little while rather than calling repeatedly. When it is FAILED, `error` says why in plain words, and nothing was charged. By default you get the summary and the transcript as plain text. Pass `include_segments` only when the person actually needs per-speaker timings, because the segment list is long. The transcript and summary are kept for a limited time after the job finishes — `content_expires_at` says until when. After that `content_expired` is true and they come back empty, so tell the person to save what they need before then. |
| list_transcripts | List this account's transcription jobs, newest first. Use it to find a job whose id the person does not remember. Segments are never included here; call `get_transcript` for one job's full result. `limit` is capped at 200. |
| get_balance | How many minutes of transcription this account can still run. Worth checking before starting a long recording. A job whose audio is longer than the balance allows is refused after the audio arrives, and is never charged. When the balance is low, tell the person to buy more minutes in their Voibe account. |