Skip to main content

MCP Integration

The Speech is Cheap Model Context Protocol (MCP) server lets AI assistants transcribe audio and video files directly within your conversations. The Model Context Protocol is an open standard that allows AI assistants to connect to external tools and data sources.

The MCP server runs every transcription request through the Jobs API using your API key. This gives your assistant access to speaker parsing, word-level timestamps, audio labels, and multiple transcript formats.

Server URL​

Connect your AI client to the Speech is Cheap MCP server at:

https://mcp.speechischeap.com/mcp

The server provides remote MCP connectivity over Streamable HTTP.

What You Need​

To connect an assistant to Speech is Cheap, you need:

  • An AI assistant or chat client that supports remote MCP connections.
  • A Speech is Cheap account with an active billing plan. Pick a plan on the Pricing page to get started. Your first API key arrives by email.
  • A Speech is Cheap API key. Keys start with sk_. You can create additional keys in the Dashboard. The dashboard displays each key only once. We recommend creating a dedicated key named for your AI client (for example, "Claude" or "ChatGPT") so you can revoke it on its own. See the Dashboard guide for details.

How Sign-In Works​

Connecting your assistant uses OAuth:

  1. When you add the MCP server, your AI app opens the Speech is Cheap sign-in page.
  2. The sign-in page displays a password field labeled Speech is Cheap API key, along with Connect and Cancel buttons.
  3. The page states that jobs are billed to the API key you enter.
  4. The server checks the key immediately:
    • If the key is incorrect or revoked, the page displays: Speech is Cheap rejected that API key.
    • If the key lacks an active subscription or billing method, the page displays: That API key has no active plan. Subscribe or add a payment method at dash.speechischeap.com.
  5. "Sign in" simply means entering your API key. There is no separate Speech is Cheap login.

The sign-in page expires after 10 minutes. If it does, start connecting again from your AI app.

Your connection remains active as long as you use it at least once every 30 days. After 30 days without use, the connection expires and you must connect again. Revoking the API key in the Dashboard ends every connection that uses it.

Available Tools and Prompts​

Every tool executes with your API key and calls the Jobs API.

Tools​

  • transcribe: Transcribes an audio or video file located at a public URL. Optional parameters include speaker parsing, word-level timestamps, audio labels, private jobs, a language code from our Supported Languages, and the output format (text, SRT, VTT, or JSON; defaults to text). The tool waits up to about 25 seconds. If the job does not finish within that window, it returns the job ID so the assistant can check status and retrieve the transcript with get_transcript.
  • get_transcript: Retrieves the status and transcript for a job ID. Long transcripts return in pages, and each page explains how to request the next page.
  • cancel_job: Cancels a pending transcription job.
  • prepare_upload: For local files, returns an upload command that reads an API key from the SIC_API_KEY environment variable. This tool is only usable by assistants that can run commands on the computer where the file resides. In chat apps such as ChatGPT and Claude, use a public URL instead.
  • check_access: Confirms that your API key works and lists current limits, file types, and prices.

Prompts​

  • transcribe: A prompt template for general audio and video transcription.
  • subtitles: A prompt template for generating subtitle files in SRT or VTT format.

Transcript Format​

In the default text format, each line starts with a timestamp, such as [m:ss], or [h:mm:ss] for long recordings. With speaker parsing, the speaker follows, such as Speaker A:. With audio labels, sounds such as music show as [music].

[0:04] Speaker A: Welcome back to the show.

Capabilities and Limitations​

Using Speech is Cheap through MCP differs from calling the Jobs API directly:

  • URLs only in chat apps: The server can't read files you attach to a chat. Give the assistant a public URL that downloads without signing in.
  • Multiple calls for long recordings: Each tool call waits at most about 25 seconds (wait_seconds, from 0 to 25). Longer recordings take more than one call: transcribe starts the job and returns a job ID, and your assistant then retrieves the transcript pages using get_transcript.
  • No webhooks or streaming: The MCP server does not support webhooks, progress event streams, or direct options for segment length, minimum confidence, and split channels.
  • Language validation: Unsupported language codes are validated and rejected before a job starts.
  • Jobs API only: The MCP server runs jobs exclusively through the Jobs API. It does not provide Edge API or x402 options.

Billing​

Billing for MCP jobs works the same as direct API usage:

  • Jobs bill to the plan associated with the API key you entered during sign-in.
  • Charges apply per started minute of audio, plus any selected add-ons.
  • Only completed jobs are billed. If a job fails, you are not charged.
  • Speech is Cheap charges no extra fee or surcharge for using MCP.
  • Visit Pricing for current rates.

What the Server Keeps​

For each active connection, the server stores:

  • Your Speech is Cheap API key (encrypted)
  • Sign-in and session tokens

The server does not store audio files, finished transcripts, or job IDs.

Disconnecting​

To stop using the integration:

  • Remove Speech is Cheap from your AI app.
  • Removing it may not delete the API key the server stores for that connection. To cut off access everywhere at once, revoke the key in the Dashboard. Revocation can take a minute or more to apply everywhere.

Client Guides​

ClientGuideDescription
ChatGPTChatGPTAdd Speech is Cheap as a custom MCP server, saved as a plugin.
ClaudeClaudeAdd Speech is Cheap as a custom connector.

Related Pages​