Works inside

Transcribe audio and video inside Claude & ChatGPT

Connect Claude, ChatGPT, Claude Code, Codex or Cursor to Whisper Web once. Ask for a transcript in plain language and get text, timestamps or subtitles back in the chat, in 99 languages.

https://whisperweb.dev/mcp

Connect your assistant

Works with any MCP client that supports Streamable HTTP and OAuth.

OAuth 2.1 · PKCE S256
10 tools
$0.01 / audio min
No API key to copy

Illustrative example · tool names match the live server

01 / Clients

Connect your client.

Every client reads the same discovery documents at /.well-known/oauth-protected-resource/mcp. Web clients call back to their own domain; CLI clients call back to localhost.

Every client uses the same URL and your normal Whisper Web sign-in.

Claude

Connector

Claude reads the OAuth metadata itself. Nothing to install.

Open Settings → Connectors → Add custom connector.
Paste the server URL and continue.
Sign in to Whisper Web and approve access.

Server URL

https://whisperweb.dev/mcp

Endpoint: whisperweb.dev/mcp
Transport: Streamable HTTP
Auth: OAuth 2.1 · PKCE S256
Default scopes: read · write · usage

ChatGPT

App

Add Whisper Web as a custom app with OAuth.

Open Settings → Apps → Advanced settings and turn on Developer mode.
Create app, paste the server URL and choose OAuth.
Sign in to Whisper Web and approve access.

Server URL

https://whisperweb.dev/mcp

Endpoint: whisperweb.dev/mcp
Transport: Streamable HTTP
Auth: OAuth 2.1 · PKCE S256
Default scopes: read · write · usage

Claude Code

Remote MCP

Signs in through a browser window and a localhost callback.

Run the command below in your terminal.
Inside Claude Code, run /mcp and choose whisperweb.
Sign in in the browser window that opens.

Terminal

claude mcp add --transport http whisperweb https://whisperweb.dev/mcp

Endpoint: whisperweb.dev/mcp
Transport: Streamable HTTP
Auth: OAuth 2.1 · PKCE S256
Default scopes: read · write · usage

Codex

Remote MCP

Signs in through a browser window and a localhost callback.

Add the server with the first command.
Run the login command.
Sign in in the browser window that opens.

Terminal

codex mcp add whisperweb --url https://whisperweb.dev/mcp
codex mcp login whisperweb

Endpoint: whisperweb.dev/mcp
Transport: Streamable HTTP
Auth: OAuth 2.1 · PKCE S256
Default scopes: read · write · usage

Cursor

mcp.json

Streamable HTTP. Cursor opens a browser window to sign in.

Add the server to your mcp.json and save.
Connect the whisperweb server.
Sign in to Whisper Web in the browser window that opens.

mcp.json

{
  "mcpServers": {
    "whisperweb": {
      "url": "https://whisperweb.dev/mcp"
    }
  }
}

Endpoint: whisperweb.dev/mcp
Transport: Streamable HTTP
Auth: OAuth 2.1 · PKCE S256
Default scopes: read · write · usage

Any MCP client

Streamable HTTP

Works with any client that supports remote MCP with OAuth discovery.

Add the URL as a Streamable HTTP server.
The client discovers OAuth from /.well-known/oauth-protected-resource/mcp.
Sign in to Whisper Web and approve access.

Server URL

https://whisperweb.dev/mcp

Endpoint: whisperweb.dev/mcp
Transport: Streamable HTTP
Auth: OAuth 2.1 · PKCE S256
Default scopes: read · write · usage

MCP needs a Developer account with purchased API credits. Add credits in the Developer Console.

Step-by-step guides: Whisper Web for Claude · Whisper Web for ChatGPT.

02 / What it does

From a link or a file to subtitles, without leaving the chat.

In

Media URLs
Direct audio or video links, plus single YouTube, Facebook and X video pages.
In-chat uploader
Pick files right in the conversation on clients that support MCP Apps.
Local files
Coding agents upload through a signed URL. Up to 5 GB and 10 hours.

Process

spoken languages

Detect the language automatically, name it, or translate the speech into English at the same rate.

validating → queued → processing → completed

Out

TXT
Segments
SRT
VTT
JSON

Long transcripts come back in pages, so the assistant can quote, summarize or rewrite them in the same conversation.

Just ask

“Transcribe this podcast and pull the key quotes with timestamps.”
“Turn my lecture recording into VTT subtitles.”
“Transcribe this Spanish interview and translate it to English.”
“How much would a 90-minute recording cost?”
“Show my last five transcriptions and open the one from yesterday.”
“Upload these two meeting recordings and summarize each one.”

03 / Toolstools/list

Ten tools. One balance.

The same Developer account, credits and records as the REST API. You never name a tool: describe what you want and the assistant picks.

get_account
Balance, pricing and limits
Free
estimate_cost
Quote for an audio length
Free
transcribe_url
Start a task from a media URL or a YouTube, Facebook or X video link
Per second
create_upload
Signed upload URL for a local file
Free
transcribe_upload
Start a task from a finished upload
Per second
start_upload
In-chat file uploader (MCP Apps)
Per second
get_transcription
Task status, duration and cost
Free
get_transcript
Text, timestamped segments, SRT, VTT or JSON, paged
Free
list_transcriptions
Recent tasks
Free
delete_transcription
Delete a finished result (separate permission)
Free

04 / Pricing

$0.01 per audio minute.

Billed by the second from prepaid API credits. Connecting, estimating, checking status and reading results are free.

starter
$10
1,000 audio minutes
Buy $10 credits
growth
$25
2,500 audio minutes
Buy $25 credits
scale
$100
10,000 audio minutes
Buy $100 credits

One-time packs. No subscription, automatic top-ups or overage charges. Applicable taxes appear at checkout. Website Unlimited is a separate plan and does not include API credits.

05 / Privacy & control

You decide what the assistant can do.

01
Sign in, no keys
OAuth 2.1 with PKCE. Access tokens are bound to whisperweb.dev/mcp and expire after an hour.
02
Deletion is opt-in
Reading, transcribing and usage are granted by default. Deleting results needs its own permission.
03
Disconnect anytime
See every connected agent in Developer Console → AI agents (MCP). After you disconnect one, its next call is refused.
04
Uploads go to Whisper Web
The in-chat uploader sends files from your browser to Whisper Web storage, not through the assistant. Transcription uses cloud processing.

Only transcribe media you own or are authorized to use. Read the API data guide.

06 / FAQ

Whisper MCP FAQ

The details that decide whether it fits.

What is the Whisper Web MCP server?

It is a remote Model Context Protocol (MCP) server at https://whisperweb.dev/mcp. MCP is the open standard that AI assistants such as Claude and ChatGPT use to work with outside tools. Once connected, your assistant can transcribe audio and video, check costs and read transcripts with your Whisper Web Developer account.

Which AI clients can I use?

Claude (web and desktop), ChatGPT, Claude Code, Codex and Cursor, plus other clients that support remote MCP servers over Streamable HTTP with OAuth. Add it as a custom connector or MCP server using the setup steps on this page.

Do I need an API key?

No. You sign in with your Whisper Web account through OAuth when the client first connects. If you build your own integration instead, use the REST API with an API key.

What does it cost?

MCP uses the same Developer API credits as the REST API: $0.01 per audio minute, billed by the second. Credit packs are $10 for 1,000 minutes, $25 for 2,500 minutes, $100 for 10,000 minutes. MCP is available to Developer accounts that have bought API credits. Checking status, estimating cost and reading results are free.

Is a website Unlimited subscription enough?

No. MCP uses Developer API credits, which are separate from the website's Free and Unlimited plans. Open the Developer Console with your Whisper Web login and buy a credit pack to start.

Can it transcribe YouTube videos?

Yes. Give your assistant a single YouTube, Facebook or X (Twitter) video link and it will transcribe the audio. Only transcribe media you own or are authorized to use, and respect the platform's terms and the creator's rights.

How do I transcribe a file on my computer?

In Claude and other clients that support MCP Apps, ask to upload a file and pick it in the in-chat uploader. Coding agents such as Claude Code, Codex and Cursor can create a signed upload URL and upload the file for you. Each file can be up to 5 GB and 10 hours.

Does my audio go to Anthropic or OpenAI?

Files picked in the in-chat uploader go from your browser to Whisper Web storage, not through the assistant. Transcription runs in Whisper Web's cloud. Your assistant receives the transcript text it requests, which then becomes part of your conversation under that assistant's own terms.

How do I disconnect an assistant?

Open Developer Console → AI agents (MCP) and choose Disconnect next to the client. Its next call is refused. You can also remove the connector inside the client.

Add credits once, then connect your client.

Open Developer Console Read the setup guide

MCP setup guide
Connection steps for each client, the full tool list and how credits are charged.
Manage connected agents
Copy the server URL, check your balance and disconnect agents in the Developer Console.
Whisper API for developers
Build your own integration with API keys, async tasks and signed webhooks.
API credit pricing
$0.01 per audio minute with one-time credit packs and no automatic top-ups.

Transcribe audio and video inside Claude & ChatGPT

Connect your client.

✳Claude

✳ChatGPT

✳Claude Code

✳Codex

✳Cursor

✳Any MCP client

From a link or a file to subtitles, without leaving the chat.

In

Process

Out

Just ask

Ten tools. One balance.

$0.01 per audio minute.

You decide what the assistant can do.

Sign in, no keys

Deletion is opt-in

Disconnect anytime

Uploads go to Whisper Web

Whisper MCP FAQ

Add credits once, then connect your client.

Claude

ChatGPT

Claude Code

Codex

Cursor

Any MCP client