Transcribe Audio and Video for LLM Interaction

Transcribe MCP connects an AI assistant (Claude, Windsurf, Cursor, and others) to a Transcribe.com account so it can automate audio transcription tasks directly. It's designed to be fast, lightweight, and LLM-friendly - no local ASR models to install or Python packages to fight with, with results returned in seconds. Transcription quality covers noisy audio, over 100 supported languages, word-level timestamps, and speaker separation, with a wide variety of input formats supported out of the box, cloud storage for audio notes and records, and team-based collaboration via Transcribe.com's teams feature. The Local Server variant can add local files (by filesystem path) for transcription, in addition to remote URLs, and return the result to the assistant.

Four tools are exposed, with more added over time: convert-to-text converts audio to text and returns it immediately, consuming account time credits - the remote server requires a publicly-accessible URL, while the Local Server accepts either a URL or a local filesystem path; get-balance returns the account's remaining balance; read-transcriptions returns the content of already-completed transcriptions with optional filtering or search; and update-transcription renames or deletes transcriptions stored in the account.

Use this when an AI assistant needs to transcribe audio - from a public URL or a local file - directly as part of a workflow, without setting up local speech-recognition infrastructure.

The private MCP integration URL (obtained from the Transcribe.com online editor's Automation popup) must be stored securely and never shared, since it authenticates the connection to the account. convert-to-text consumes account time credits per use, so balance should be checked with get-balance before running large batches.

Input is an audio file, either as a publicly-accessible URL (required for the remote server) or a local filesystem path (Local Server only). Output is the transcribed text, with word-level timestamps and speaker separation where applicable, retrievable later via read-transcriptions.

For Claude Desktop, sign in to the Transcribe online editor to get a private MCP integration URL, then download the pre-built MCPB bundle and either double-click it for one-click installation or install it manually via Settings > Extensions > Install Extension, pasting the private URL when prompted. For other assistants, install Node.js and Git first (verify with node --version and npx --version), then add the Local Server via your assistant's settings using the server config snippet - note the source's own example snippet nests an args key inside another args array, which looks like a documentation typo rather than valid JSON; the intent is to run npx -y github:transcribe-app/mcp-transcribe with MCP_INTEGRATION_URL set to your private URL as an environment variable. If installation fails with a missing MCP_INTEGRATION_URL error, re-copy the private URL from the Automation popup and set it as that environment variable; if the extension won't install, confirm Claude Desktop is up to date and that node is on your PATH.

Users who want an AI assistant to transcribe audio - from URLs or local files - directly within a conversation, without running local speech-recognition models.

{
  "key": "transcribe-local",
  "command": "npx",
  "args": [
    "args": ["-y", "github:transcribe-app/mcp-transcribe"],
  ],
  "env": {
    "MCP_INTEGRATION_URL": "<your-MCP-integration-URL>"
  }
}