Whisper API and MCP: how to add speech recognition to your project

Whisper is a speech-to-text model. NeuralSpace offers the following models: Whisper large-v3, Whisper. They all run through a single public API, and there is a ready MCP server for AI agents. The key is issued on the API keys page. Below are the model capabilities, request examples and setup steps.

What Whisper can do

Whisper transcribes speech from an audio file and returns text. The request is sent as multipart/form-data with a file field and the whisper-large-v3 model (the whisper-1 id transparently works as large-v3). mp3, mp4, mpeg, mpga, m4a, wav, webm, ogg and flac are supported, up to 25 MB. The price is per minute of audio.

API endpoint

Send a request to POST /v1/audio/transcriptions. Example: multipart/form-data: file (audio), model=whisper-large-v3. Authorization uses an API key in the x-api-key header. Available model ids: whisper-large-v3, whisper-1.

MCP tools

The NeuralSpace MCP server is available at POST /mcp (Streamable HTTP). To work with Whisper, an agent uses the tools: transcribe_audio, list_models, get_balance. This lets Claude Code, Cursor and other MCP clients use the site models directly.

How to connect

Sign up on NeuralSpace, open the API keys page and create an nsk-… key. Then pass it in the request header or in your MCP client settings. Full documentation is on the the /v1/docs page.

FAQ

Which audio formats are supported?

mp3, mp4, mpeg, mpga, m4a, wav, webm, ogg and flac.

What is the maximum file size?

25 MB.

What format is the request?

POST /v1/audio/transcriptions as multipart/form-data with a file field.