ElevenLabs MCP: Introduction
Learn how to connect the ElevenLabs MCP to ChatGPT or Claude to generate and manage transcripts with speaker labels and timestamps.
Last updated
Menus and supported features can change. Compare the source-check date below with your current app version.

Connecting transcription tasks with ElevenLabs MCP
This reader-requested task follows ElevenLabs MCP official documentation, checked on 2026-11-07. The exact first release date and deployed build are unverified; selection does not imply a recent launch. Documentation edit dates are not release dates. This is a documentation-based guide, not a hands-on test or a paid purchase.
If working with audio files has you switching between windows, connecting the task to an AI assistant may be worth considering. ElevenLabs MCP is a hosted server connecting an ElevenLabs workspace to an AI assistant F1. The official announcement describes requesting creative tasks such as speech, transcription and dubbing through clients including ChatGPT and Claude F2. This introduction focuses on transcribing one short English recording, explaining how the connection works and what to check.
The documentation says there is no need to run a separate local server. You sign in with an ElevenLabs account and use OAuth to grant scoped workspace access, without copying an API key into the client F3. Check the requested permissions and the workspace you intend to use F13. Prepare a recording you have permission to send to the service.
Check your environment and costs first
The announcement lists supported clients including Claude, ChatGPT and Cursor F2. The ChatGPT instructions describe adding ElevenLabs from the plugins directory and signing in, with an OAuth connection that can be revoked later F6. The Claude instructions describe adding a custom connection through Customize → Connectors using the official URL F7. Read this as a separate workflow from adding external MCP servers to an ElevenLabs voice agent.
Do not assume every client exposes the same capabilities. For example, the documentation describes the ElevenLabs connector in Claude’s connector directory as a restricted version without image, video or music generation F5. Even if a connection is visible, first check whether it actually exposes transcription and file transfer tools. Account-specific availability, exact prices, credit usage and commercial-use conditions have not been verified for this guide. Before running a task, check the official pricing page, terms and your account’s actual billing conditions D4, D5. A successful connection does not establish that the task is free.
Documented transcription features and remaining checks
The official creative announcement says that passing a recording to Scribe returns a transcript with speaker labels and timestamps F4. This is a documented capability, not a result obtained by running the example below. Before making a request, inspect the connected assistant’s actual transcription input format and returned fields. If no attachment or transcription tool is available, stop and consider a route supported by official documentation, such as the web workflow.
The web Speech to Text documentation identifies Scribe v2 as the website’s default model F12. That alone does not establish that a particular MCP client calls the same model. If the actual tool description does not name a model, record “MCP model unverified.” MCP export buttons and advanced options have not been established from these materials either. Do not present the web editor’s menus as MCP features.
Shared example input and one request
The following is an editorial example prepared for this series. It is not an official example or a supplied file: the reader must prepare an actual recording they have permission to make and use. No recording, transcription or cost verification was performed for this guide. Once you have a file containing the two speakers’ short sentences, check that the filename and description in the request match the actual input.
Owned input: two-speakers.wav, a 20-second English recording made with permission. Speaker A says: The meeting starts at ten. Speaker B says: I will bring the notes. Request: Transcribe this file once, preserve the words, show speaker labels and available timestamps, and mark uncertain words instead of guessing.
Confirm that the connected tool can accept the input, then attach the actual file once and make one request. The expected result is text corresponding to the recording and whatever speaker and timing information the tool actually returns F4. If the result is pending, check the original task’s status rather than resubmitting the same file because it is delayed. Asking the assistant not to invent missing timing information or inaudible words is also an editorial verification suggestion.
Save the result and compare it with the original
If text has been returned, you can paste it into a normal text editor and save it as owned-transcript.txt in UTF-8. This is an editorial suggestion for manual file preparation, not a built-in MCP export feature. Keep timing JSON alongside it only if such JSON was actually returned; do not assume it exists otherwise.
Reopen the saved file and compare the two speakers’ words and speaker changes with the original audio. Leave unclear sections marked uncertain, and compare any returned timestamps with playback positions. Plausible wording is not the same as a faithful transcription, so the final file still needs review. This guide is not a report of measured accuracy, processing time or tested file creation.
Compare with the web workflow before choosing
In web Speech to Text, Transcribe files lets you upload audio or video and select a language or leave it on Detect F9. The instructions describe opening results by clicking the uploaded filename, clicking a word to play audio at that point, and downloading through Export F10. Plain text, JSON, HTML, SRT and VTT are documented web Transcript Editor export options F11. Whether an MCP job appears in web history or offers the same export menus requires separate verification.
As an editorial judgment, MCP may be worth considering for people who want to connect tasks within a conversation, while the web workflow offers selection criteria for people interested in word-level playback and on-screen editing. Output, time and cost have not been measured using the same input, so this guide does not conclude that either route is faster or more accurate. Check actual client compatibility, file transfer friction, the capabilities needed for automation and the intended audience before choosing the workflow that suits your needs.
Original BumbleV reading guide illustration; not a product screenshot or a result from our testing. Official release behind this illustration · CC BY 4.0 · 2026-11-06

Sources & further reading
- https://elevenlabs.io/blog/introducing-voice-music-image-and-video-generation-in-the-elevenlabs-mcp
- https://elevenlabs.io/mcp
- https://elevenlabs.io/docs/eleven-agents/operate/hosted-mcp
- https://elevenlabs.io/pricing
- https://elevenlabs.io/terms-of-use
- https://elevenlabs.io/docs/eleven-creative/playground/speech-to-text
- https://elevenlabs.io/docs/eleven-creative/products/transcripts