TollNode

Speech to text API and MCP for AI agents

TollNode gives Claude, Cursor or any AI agent a transcript of any MP3 or WAV link, with punctuation, in 50+ languages and up to 25 MB, from $0.015 per clip, with no subscription.

Ask in plain words, like 'transcribe this meeting recording and list the action items', and get the text back.

No subscription · Sign-up optional · From $0.015 per call · Charged only when it works · No OpenAI account needed

Claude Code

Transcribe this recording: tollnode.com/examples/elevenlabs-turbo-speech.mp3

TollNode · OpenAI · 1$0.015 · 1 clip

ResultDetail
TranscriptGood morning. Here is your briefing: three meetings today, light rain until noon...
Languageen
FileMP3 under 1 minute

Real result.

Works in Claude Code, Cursor, Codex, VS Code and any MCP agent.

What your agent gets from a transcription

Every result is structured JSON your agent can sort, filter, summarize or turn into a spreadsheet.

What people ask their AI to do with it

Use it as a Speech to text MCP server in Claude, Cursor or Codex

Connect once. Then just ask.

Paste this into Claude Code, Cursor or Codex. It connects TollNode and gives your agent a small wallet. Add a few dollars of USDC and you're done.

terminal
npx agentcash add https://tollnode.com

Copy for your AI agent

Paste this into your agent so it knows what this node does and how to call it.

prompt
You can use TollNode's Speech to text node. Call POST https://tollnode.com/api/v1/nodes/speech-to-text with audio (required), prompt (optional), language (optional), max_minutes (optional). The price depends on max_minutes (1, 5, 15, 30; default 5) and is charged only when a result is returned. Over MCP, call the "speech-to-text" tool; use list_models to see prices. Return results as tables when I ask for lists.

Price

from $0.014

per minute

One call: from $0.015 per transcript up to 5 minutes

The price depends on the max_minutes you choose. If you don't choose, you get up to 5 minutes.

up to 1 minute$0.015 per recording up to 1 minute
up to 5 minutesdefault$0.069 per recording up to 5 minutes
up to 15 minutes$0.204 per recording up to 15 minutes
up to 30 minutes$0.406 per recording up to 30 minutes

Charged only when a call returns a result. Failed or empty calls cost $0.

Speech to text without running your own model

FAQ

Is this the official OpenAI API?
No. TollNode is an independent service. It gives your agent speech-to-text through its own API and MCP server, priced per clip by the longest recording you allow.
Can I use it as a Speech to text MCP server in Claude?
Yes. Add https://tollnode.com/mcp in Claude Code, Claude Desktop or Cursor, and the "speech-to-text" tool appears with its fields and price.
Which files work?
A public link to an MP3 or WAV file, up to 25 MB. M4A and video files are not accepted.
What does max_minutes do?
It sets the longest recording you are willing to pay for: 1, 5, 15 or 30 minutes. Longer files are refused before anything runs and are not charged.
How many languages?
More than 50. Send a 2-letter language code, such as en, to improve accuracy.
Do I need a OpenAI account?
No. Your agent calls TollNode, and you never log in to OpenAI or share cookies.
How do I pay for Speech to text?
Your agent sends the request, gets an HTTP 402 with the exact price (from $0.015 per call), signs a USDC payment on Base and retries. Or use an account balance with an API key. Either way you are charged only when a result comes back.
What does the Speech to text response look like?
Structured JSON, with the fields listed in the cards above. Your agent can sort it, filter it, summarize it or turn it into a table or spreadsheet.

Ready in a minute

Connect your AI
For developers: endpoint, inputs and an example response

POST https://tollnode.com/api/v1/nodes/speech-to-text

Input

FieldTypeNotes
audio*stringPublic URL of the recording as MP3 or WAV, up to 25 MB (M4A and video files are not accepted). Or send the file inline as a base64 data: URI (up to 4.5 MB)
promptstringNames, terms or context that help spell things right. Optional
languagestringLanguage of the speech as a 2-letter code, e.g. en; improves accuracy. Optional
max_minutesenumLongest the recording can be; sets the price. Longer files are refused before anything runs, and not charged. Default 5 (1, 5, 15, 30)

Example response

{
  "transcript": "Giles. Part 1. Laying Plans. Sun Tzu said, The art of war is of vital importance to the state…"
}

Also described in the OpenAPI file and llms.txt. Setup for every app: Agents & MCP.