Menu
Mattering Automations connector

An ElevenLabs MCP that makes voice a workflow step, not a separate tool

Paste one API key and ElevenLabs text to speech, transcription, dubbing, sound effects, the public voice library, and the Conversational AI agents platform become nodes for visual automations and AI agents. The audio comes back inside the workflow, ready to land in a task, a database, or the next step, and every node that spends credits says so in its own description.

37 ready-made nodes One credential to connect Works with Claude, ChatGPT, and Matter

What the ElevenLabs MCP connector does

The ElevenLabs connector for Mattering Automations is 37 nodes curated from ElevenLabs' REST API: text to speech in both binary and JSON shapes, sound effect generation, Scribe transcription from a URL, the full asynchronous dubbing lifecycle, voice management plus search of the public voice library, pronunciation dictionaries built from JSON rules, the models and account plane for picking a model and watching credits, generation history, and the Conversational AI agents platform with its conversation logs. It is a curated surface rather than a wrapper around everything, and the parts left out are named plainly further down.

Every node speaks MCP, the open protocol for connecting AI assistants to tools, so Claude, ChatGPT, and any MCP-capable assistant can use them, alongside Matter, the AI assistant built into the platform. ElevenLabs also publishes an official MCP server of its own, and this page says so out loud; the difference is where the nodes live. Here the same operations sit on an automation canvas next to hundreds of other connectors, run on schedules and webhooks with nobody at a keyboard, and hand their output to whatever comes next.

Because Mattering is a work platform rather than middleware, voice work lands somewhere useful by default: a narrated version of an article can attach to the task that requested it, a call transcript from a Conversational AI agent can become a QA row in a Mattering database, and a weekly credit check can message a channel before a batch job runs the account dry.

Four things are worth knowing before you wire anything. The API key travels in a custom xi-api-key header, not an Authorization Bearer header, and the connector formats that for you. Audio goes in as a URL that ElevenLabs fetches itself, never as an upload. Audio comes back as base64 inside the node output, under a 5 MB encoded cap, with ElevenLabs: Text to Speech with Timestamps as the JSON-native alternative when downstream steps consume the result. And billing is a credit pool with concurrency-based limits per plan rather than requests per minute; no rate is quoted anywhere on this page on purpose, because rates change and elevenlabs.io/pricing is the source.

One connection, three surfaces

  • Visual automations. Drag ElevenLabs nodes onto the canvas and wire them to triggers, conditions, and other apps.
  • AI agents, yours or ours. These nodes speak MCP, the open protocol, so Claude, ChatGPT, and any MCP-capable assistant can use them, and Matter, the agent built into the platform, uses them natively.
  • The whole platform. Combine with native tasks, projects, and other connectors, no middleware in between.

Every ElevenLabs node, ready to drop on the canvas

37 nodes, generated from the connector itself so this list always matches what ships. Expand a category to browse.

Read and query nodes 21

ElevenLabs: Get Agent

Get one Conversational AI agent's full configuration: prompt, first message, language, voice and conversation settings.

ElevenLabs: Get Character Usage

Get credit usage over a time window as parallel time and usage arrays, optionally broken down by voice, model, API key or user and bucketed by hour, day, week or month. Note: the docs mark this endpoint deprecated in favor of a newer workspace analytics query, but it remains the simple documented usage read.

ElevenLabs: Get Conversation

Get one conversation's full detail: the turn-by-turn transcript with tool calls, call metadata (duration, cost) and the post-call analysis (summary, success evaluation, collected data).

ElevenLabs: Get Default Voice Settings

Get the workspace default voice settings (stability, similarity boost, style, speed, speaker boost) applied to new voices and used when a request omits voice_settings.

ElevenLabs: Get Dubbed Audio

Download the dubbed media for one language of a finished dub. Returns base64 (data.audio_base64) with content type and byte size, subject to the 5 MB encoded cap - fine for short audio dubs; full videos will exceed it and must be downloaded outside the canvas.

ElevenLabs: Get Dubbing

Get a dubbing project's status and metadata. status moves from dubbing to dubbed (or failed with an error message); poll this between Dub Media from URL and the transcript or audio nodes.

ElevenLabs: Get Dubbing Transcript

Get the transcript of a finished dub in one of its languages as subtitle text (SRT by default, or WebVTT). Returns the raw subtitle document at data.text - ready to pipe into a file or captioning step.

ElevenLabs: Get History Audio

Download the audio of a previously generated item without paying to regenerate it. Returns base64 audio (data.audio_base64) with content type and byte size, subject to the 5 MB encoded cap; longer clips exceed it and must be fetched outside the canvas. The bulk zip download endpoint (POST /v1/history/download) is not shipped for the same reason.

ElevenLabs: Get History Item

Get one generated item's metadata: the text, voice, model, date and character cost. For the audio itself use Get History Audio.

ElevenLabs: Get Pronunciation Dictionary

Get one pronunciation dictionary's metadata including its latest version ID and rule count.

ElevenLabs: Get Subscription

Get extended subscription details: tier, credits used vs the monthly limit, the reset timestamp, voice slot usage and feature entitlements. The natural guard node before a big TTS batch.

ElevenLabs: Get User Info

Get the account behind the API key: user ID, name, onboarding state and a summary of the subscription including characters used and the monthly limit.

ElevenLabs: Get Voice

Get full metadata for one voice: name, category, labels, description, preview URL, saved settings and fine-tuning state.

ElevenLabs: Get Voice Settings

Get the saved settings for one voice: stability, similarity boost, style, speed and speaker boost.

ElevenLabs: List Agents

List the workspace's Conversational AI agents with their IDs, names, tags and creation times.

ElevenLabs: List Conversations

List Conversational AI calls with duration, message count, status and success evaluation - the feed for call-log and QA automations.

ElevenLabs: List History

List previously generated audio items with their text, voice, model and character cost. Paginate with start_after_history_item_id from the previous response's last_history_item_id.

ElevenLabs: List Models

List every available model with its capabilities (text to speech, voice conversion, style support), supported languages, character caps and cost factors. Use it to pick a model_id for the speech nodes.

ElevenLabs: List Pronunciation Dictionaries

List the workspace's pronunciation dictionaries with their IDs, latest version IDs and rule counts. Dictionary id plus version_id is the locator shape TTS nodes accept in pronunciation_dictionary_locators.

ElevenLabs: List Voices

List the voices available to this account - premade, cloned, generated and library-saved - with their voice IDs, settings and metadata. Filter by search text, type or category. This is the free lookup that feeds voice_id into every speech node, and the credential health check for the connector.

ElevenLabs: Search Voice Library

Search the public ElevenLabs voice library of community and professional shared voices by text, gender, age, accent, language and use case. Each result carries the public_owner_id and voice_id needed by the Add Shared Voice node.

Action nodes 16

ElevenLabs: Add Pronunciation Rules

Add rules to an existing pronunciation dictionary, creating a new version. A rule with the same string_to_replace as an existing one replaces it. Reversible with Remove Pronunciation Rules.

ElevenLabs: Add Shared Voice

Add a voice from the public voice library to this account's voices under a name of your choosing. Uses a voice slot; reversible with the Delete Voice node. Get public_user_id and voice_id from the Search Voice Library node.

ElevenLabs: Create Agent

Create a Conversational AI agent from a conversation_config object (prompt, first message, language, voice). Returns the new agent_id; reversible with Delete Agent. Running conversations with the agent bills credits per minute.

ElevenLabs: Create Pronunciation Dictionary

Create a pronunciation dictionary from a JSON list of rules (no file upload needed). Rules are alias replacements ('AI' spoken as 'ay eye') or phoneme overrides in IPA or CMU. Returns the new dictionary id and version_id to use in TTS pronunciation_dictionary_locators.

ElevenLabs: Delete Agent

Permanently delete a Conversational AI agent and its configuration.

ElevenLabs: Delete Conversation

Permanently delete a conversation record including its transcript, analysis and audio.

ElevenLabs: Delete Dubbing

Permanently delete a dubbing project including its dubbed media and transcripts. Does not refund the credits spent.

ElevenLabs: Delete History Item

Permanently delete one generated item and its audio from the account history. Does not refund the characters spent.

ElevenLabs: Delete Voice

Permanently delete a voice from this account. Cloned and generated voices cannot be recovered; premade voices cannot be deleted. Frees a voice slot.

ElevenLabs: Dub Media from URL

Start an automatic dubbing job that translates and revoices audio or video from a URL into a target language, keeping the speakers' voice character. Returns a dubbing_id immediately; the job runs asynchronously - poll Get Dubbing until status is dubbed, then fetch the transcript or audio. Billed in credits by media duration. Direct file upload is not supported on the canvas - host the media at a URL first.

ElevenLabs: Edit Voice Settings

Update the saved settings for one voice. Omitted fields keep their current values; read them first with Get Voice Settings if you need to restore later.

ElevenLabs: Generate Sound Effect

Generate a sound effect from a text description (e.g. 'heavy rain on a tin roof'). Returns the audio as base64 (data.audio_base64) with content type and byte size, subject to the 5 MB encoded cap. Billed in credits; effects are 0.5 to 30 seconds long.

ElevenLabs: Remove Pronunciation Rules

Remove rules from a pronunciation dictionary by their string_to_replace values, creating a new version. The inverse of Add Pronunciation Rules - re-adding the same rules restores them.

ElevenLabs: Text to Speech

Convert text into speech with an ElevenLabs voice. Returns the audio as base64 (data.audio_base64) with its content type and byte size; responses over the 5 MB encoded cap fail with a clear error, so use a lower-bitrate output_format or shorter text for long passages. Billed in credits per character. For JSON output with per-character timing use the Text to Speech with Timestamps node instead.

ElevenLabs: Text to Speech with Timestamps

Convert text into speech and get a JSON response containing the audio as base64 (data.audio_base64) plus per-character start and end times in seconds (data.alignment) - the shape subtitle and karaoke-style pipelines need. Same billing as plain text to speech. This is the JSON-native TTS node, so prefer it when downstream nodes consume the output.

ElevenLabs: Transcribe Audio from URL

Transcribe an audio or video file with ElevenLabs Scribe by pointing at a public HTTPS URL (up to 2 GB) - no upload needed. Returns the transcript text with per-word timestamps, optional speaker diarization and audio-event tags. Billed by audio duration. The request is held open while the file is processed, so long media can take minutes; this node allows up to 5. Direct file upload is not supported on the canvas - host the file at a URL first.

Connecting ElevenLabs takes one credential

ElevenLabs issues API keys from the profile menu at elevenlabs.io, under API Keys. Keys can be restricted per product in the dashboard, and a key without permission for a product returns 401 with a missing_permissions message even though the key itself is valid, so create an unrestricted key or grant it the products your nodes touch: text to speech, speech to text, dubbing, sound effects, and Conversational AI as applicable. The key is sent as the custom header xi-api-key on every request, not as an Authorization Bearer header; inside this connector you never type the header, because the credential is a single secret value and the connector formats it for you.

1

Sign in at elevenlabs.io, open your profile menu, go to API Keys, and create a key. Either leave it unrestricted or grant it the products your automations will use, and copy the value.

2

In Mattering Automations, open Connectors, click Add Connector, choose ElevenLabs, and paste the key. The form asks for one value and nothing else: there is no header name to type, because the connector sends xi-api-key itself.

3

Click Test Connection. Mattering calls the ElevenLabs voices list, which is free, read-only, and works on every tier including the free plan, and a bad key fails it immediately.

4

All 37 ElevenLabs nodes appear in the canvas palette. Run ElevenLabs: List Voices first, because its voice_id values feed every speech node, then ElevenLabs: List Models to pick a model_id.

Common uses for the ElevenLabs MCP

Text to speech that hands its audio to the next step

ElevenLabs: Text to Speech converts text with any voice on the account and returns the audio as base64 at data.audio_base64, with content type and byte size beside it, billed in credits per character. The response is capped at 5 MB encoded, which at the default mp3_44100_128 output format is roughly 4 minutes of audio and at mp3_22050_32 roughly 16 minutes, so long passages either drop the bitrate or split the text. When the next node needs JSON rather than an audio blob, use ElevenLabs: Text to Speech with Timestamps instead: the vendor itself returns the base64 audio plus per-character start and end times as JSON.

Subtitles and karaoke timing from a single call

ElevenLabs: Text to Speech with Timestamps returns the audio and a character-level alignment in one response: each character of the input with its start and end time in seconds at data.alignment. That is the exact shape subtitle generation, karaoke-style highlighting, and word-synced video pipelines need, without a second transcription pass over audio the workflow just generated. It bills the same as plain text to speech, so choosing it costs nothing extra.

Transcription without uploading anything

ElevenLabs: Transcribe Audio from URL runs Scribe over an audio or video file by pointing at a public HTTPS URL of up to 2 GB, and ElevenLabs fetches the media itself. The response carries the transcript with per-word timestamps, optional speaker diarization, and audio-event tags, billed by media duration. The request is held open while the file processes, so long media can take minutes; the node allows up to 5. There is no file upload path on the canvas, so media that only exists locally goes to object storage or any fetchable URL first.

Dub a video into another language while the workflow waits politely

Dubbing is asynchronous by design: ElevenLabs: Dub Media from URL starts a job that translates and revoices media from a URL into a target language while keeping each speaker's voice character, and returns a dubbing_id immediately. ElevenLabs: Get Dubbing reads the job status as it moves from dubbing to dubbed, and once it lands, ElevenLabs: Get Dubbing Transcript returns SRT or WebVTT subtitle text and ElevenLabs: Get Dubbed Audio returns the dubbed track as base64 under the same 5 MB cap, which suits short audio dubs; full videos are downloaded outside the canvas.

Call logs and QA from Conversational AI agents

ElevenLabs: List Conversations feeds a call-log automation with every agent conversation's duration, message count, status, and success evaluation, and ElevenLabs: Get Conversation opens one of them: the turn-by-turn transcript with tool calls, call metadata including cost, and the post-call analysis with its summary and collected data. A scheduled flow can pull yesterday's calls, write the analysis into a Mattering database, and flag failed evaluations as tasks for a human, and ElevenLabs: Create Agent and ElevenLabs: Get Agent manage the agents themselves from the same canvas.

Credit guard rails before a batch job

ElevenLabs bills every product from one shared credit pool, so a batch TTS run and a dubbing job draw down the same balance. ElevenLabs: Get Subscription reads the tier, credits used against the monthly limit, and the reset timestamp, which makes it the natural condition before any loop that spends; ElevenLabs: Get Character Usage turns spend over a time window into arrays a reporting flow can chart, broken down by voice, model, or API key. Both are free reads, so they can run on a schedule without touching the meter.

Example: A published article becomes a narrated audio version, with the balance checked first

Built on the Automations canvas in a few minutes, or described in plain language to Matter and built for you.

1

Trigger: a webhook or scheduled check delivers the text of a newly published article.

2

ElevenLabs: Get Subscription reads credits used against the monthly limit, and a condition stops the flow before it spends anything if the remaining balance is below the floor you set. This step is free.

3

ElevenLabs: List Voices returns the account's voices; the flow picks the narration voice_id, either fixed in configuration or chosen by an agent from the voice metadata. Also free.

4

ElevenLabs: Text to Speech converts the article text with that voice. This is the step that costs credits, billed per character, and its output is the audio as base64 with its content type and size. For a long article, set a lower-bitrate output_format so the response stays under the 5 MB encoded cap.

5

The audio lands where the work is: attached to the article's task in Mattering Productivity, stored via a file step, or handed to a distribution workflow.

6

A weekly ElevenLabs: Get Character Usage on the same account turns the flow's spend into a number the team can see without opening the ElevenLabs dashboard.

ElevenLabs MCP questions, answered

What ElevenLabs operations are supported?

37 nodes: text to speech in binary and JSON-with-timestamps shapes, sound effect generation, Scribe transcription from a URL, the dubbing lifecycle from job creation through transcript and audio download, voice management including public voice library search and adding shared voices, pronunciation dictionaries created and edited from JSON rules, models, account, subscription and usage reads, generation history with audio re-download, and Conversational AI agents with their conversation logs. Deliberately not shipped: voice cloning, every streaming and WebSocket endpoint, file-upload-only endpoints such as audio isolation and forced alignment, the Studio and Music surfaces, agent update, and workspace administration. Five nodes are flagged destructive: ElevenLabs: Delete Voice, ElevenLabs: Delete History Item, ElevenLabs: Delete Dubbing, ElevenLabs: Delete Agent, and ElevenLabs: Delete Conversation.

ElevenLabs already ships an official MCP server. Why use this one?

It does, and it is good: the official elevenlabs-mcp on GitHub gives a single assistant direct ElevenLabs tools, and if all you want is Claude talking to ElevenLabs from your own machine, it is the shorter path. This connector earns its keep when voice is one step in a bigger system. The same 37 nodes run in visual automations on schedules and webhooks with no assistant in the loop, chain with the rest of Mattering's connectors, and drop their output into tasks, databases, and messages where a team actually sees it. Any MCP-capable assistant can drive them too, so choosing this is not choosing against MCP; it is choosing where the work lands.

Does this work with Claude, ChatGPT, or other AI assistants?

Yes. Mattering exposes these nodes over MCP, an open protocol for connecting AI assistants to tools, so any MCP-capable assistant, including Claude and ChatGPT, can connect to your workspace and use them. Matter, the assistant built into the platform, uses the same nodes natively. The read nodes such as voices, models, history, and conversation logs are free and side-effect-free, which makes them the safest surface to hand an agent; the generation nodes spend credits on every run, so gate those the way you would any spending step.

Why does a valid ElevenLabs API key return 401?

Two usual causes. First, the header: ElevenLabs authenticates with the custom header xi-api-key, not Authorization Bearer, so a Bearer header copied from another AI API fails with a perfectly good key. Inside this connector the header is formatted for you, so that mistake is only possible from a generic HTTP node. Second, key restrictions: ElevenLabs keys can be scoped per product in the dashboard, and a key without the relevant permission returns 401 with a missing_permissions message. The connection test uses the voices list, which most keys can read, so a scoped key can pass Test Connection and still 401 on, say, dubbing; grant the key the products your nodes touch.

Can I upload an audio or video file?

No. Media reaches ElevenLabs as a URL that ElevenLabs fetches itself: ElevenLabs: Transcribe Audio from URL takes a public HTTPS link of up to 2 GB, and ElevenLabs: Dub Media from URL takes the source the same way. An automation canvas cannot produce a file-upload request body, so the endpoints that only accept uploaded files, such as audio isolation and forced alignment, are not shipped at all rather than shipped broken. In practice a public link works and so does a signed, time-limited URL from object storage, as long as it is still fetchable at the moment the node runs.

How does generated audio come back, and what is the size limit?

As base64 inside the node's JSON output: data.audio_base64 with a content type and byte size beside it, on ElevenLabs: Text to Speech, ElevenLabs: Generate Sound Effect, ElevenLabs: Get History Audio, and ElevenLabs: Get Dubbed Audio. The encoded payload is capped at 5 MB, and a response over the cap fails with a clear error instead of a truncated blob. At the default mp3_44100_128 output format that is roughly 4 minutes of audio; at mp3_22050_32 roughly 16 minutes. For long passages, lower the bitrate or split the text, and when downstream nodes consume the result, prefer ElevenLabs: Text to Speech with Timestamps, whose JSON response comes from ElevenLabs itself. Full dubbed videos exceed the cap and are downloaded outside the canvas.

Can I clone my voice through this connector?

No, and that is deliberate rather than missing. ElevenLabs' terms require consent for cloning a voice, and that consent is a per-sample judgement a human should make; putting instant voice cloning one node away from an automation loop is the wrong default, so the cloning endpoints are not shipped. What is shipped is the consent-safe path: ElevenLabs: Search Voice Library finds community and professional voices already published to the public library under their owners' terms, and ElevenLabs: Add Shared Voice saves one to your account for use in any speech node. Voices you have already cloned in the ElevenLabs dashboard appear in ElevenLabs: List Voices and work normally.

What does it cost, and is there a free tier?

ElevenLabs bills everything from one monthly credit pool: text to speech per character, transcription and dubbing by media duration, Conversational AI per minute. The free plan includes API access with 10,000 credits per month, which is enough to test every node family in this connector, but free-tier output carries no commercial license and may be used only non-commercially with attribution to ElevenLabs; commercial licensing starts at the Starter plan, and elevenlabs.io/pricing is the source for current plans and rates, none of which are quoted here because they change. Limits are concurrency-based per plan rather than requests per minute, so a batch flow that fans out too wide gets 429 responses, which this connector surfaces as failed step outputs rather than crashes.

37 nodes waiting on the canvas

Put ElevenLabs to work inside your workspace

Connect once, then use every node in visual automations and AI agent workflows across the Mattering platform.

All product names, logos, and brands are property of their respective owners and are used for identification purposes only. Use of these names, logos, and brands does not imply endorsement or affiliation. ElevenLabs is a product of its respective owner; Mattering is an independent platform.

Start Free Explore our Products