Menu
Mattering Automations connector

An AssemblyAI MCP for the workflow after the transcript

Connect one API key and transcription becomes a workflow step instead of a destination. Submit a recording by URL, poll the job to completion, and then keep going: pull the action items with their timestamps, export subtitles, search the call for the words that matter, and land the results next to your tasks and databases instead of in a downloads folder.

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

What the AssemblyAI MCP connector does

The AssemblyAI connector for Mattering Automations is 14 nodes across AssemblyAI's async speech-to-text API and its LLM Gateway: submit audio or video for transcription by URL, poll the job to completion, read the finished transcript as sentences, paragraphs, or SRT and VTT subtitles, search it for specific words and phrases, fetch the PII-redacted audio file, and run summaries, action items, translations, and chat-style questions over the result. One API key authenticates all of it.

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. An agent can submit a meeting recording, wait for the transcript, pull each action item with its supporting quote and timestamp, and open a task for every one, without anyone pasting text between tools.

AssemblyAI publishes its own official MCP server, and it is good at what it is for: giving an AI coding tool a direct line to the AssemblyAI API from inside your editor. This connector is for the other half of the job, where transcription is one step among many. On the Mattering canvas the same nodes sit next to every other connector and the platform itself, so a new recording can become a transcript, a summary, and a set of follow-up tasks in one run, a word search hit can raise an alert in another system, and the results can be written into a Mattering database. Use whichever fits, or both.

The safety modeling is stated plainly. Of the 14 nodes, seven read and seven act, and exactly one is flagged destructive: AssemblyAI: Delete Transcript permanently removes a transcript's text and derived data from AssemblyAI. Two surfaces are deliberately out of scope: streaming speech-to-text runs over WebSocket, which is not a canvas transport, and raw file upload is not shipped because the submit node takes a URL natively and AssemblyAI downloads the file itself. LeMUR, AssemblyAI's earlier LLM framework, was retired on March 31, 2026, so this connector ships its replacement instead: LLM Gateway chat completions and the speech understanding nodes.

One connection, three surfaces

  • Visual automations. Drag AssemblyAI 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 AssemblyAI node, ready to drop on the canvas

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

Read and query nodes 7

AssemblyAI: Get Redacted Audio

Fetch the download URL for a transcript's PII-redacted audio file (the version with sensitive speech beeped out). Only exists when the transcript was submitted with redact_pii_audio true, and the returned URL expires 24 hours after generation, so download it promptly downstream.

AssemblyAI: Get Subtitles

Export a completed transcript as subtitle text in SRT or VTT format, returned verbatim at data.text (capped at 200,000 characters with a truncation flag). Pipe it into a file step or a video pipeline downstream.

AssemblyAI: Get Transcript

Fetch a transcript job by id: its status plus, once completed, the full text, word-level timings and confidences, and any enabled intelligence results (highlights, sentiment, entities, content safety). Status moves queued to processing to completed or error, so this is the polling read after AssemblyAI: Submit Transcript: re-run it until data.status is completed, then branch on the result.

AssemblyAI: Get Transcript Paragraphs

Fetch a completed transcript semantically segmented into reader-friendly paragraphs, each with start/end times, a confidence score and word-level detail. The right shape for publishing a transcript as readable text.

AssemblyAI: Get Transcript Sentences

Fetch a completed transcript segmented into sentences, each with start/end times in milliseconds, a confidence score, the speaker (when diarization was enabled) and word-level detail. The natural unit for clipping, quoting and sentence-level analysis downstream.

AssemblyAI: List Transcripts

List the account's transcript jobs, newest first, with each job's id, status, source audio URL and timestamps. Filter by status or creation date; page with before_id/after_id cursors (data.page_details carries ready-made prev_url and next_url). The transcript IDs from here feed every other AssemblyAI node, and this is also the cheapest credential health check to run first in any AssemblyAI automation.

AssemblyAI: Word Search

Search a completed transcript for words, numbers or short phrases (up to five words each). Returns each match with its occurrence count, millisecond timestamps and word indexes, so an automation can jump straight to the moments that matter (competitor mentions, pricing talk, names).

Action nodes 7

AssemblyAI: Delete Transcript

DESTRUCTIVE: permanently remove a transcript's data from AssemblyAI. The transcript text, source audio URL and all derived results are irrecoverably deleted; the id remains but returns placeholder values. Capture anything you need with AssemblyAI: Get Transcript first.

AssemblyAI: Extract Action Items

Extract action items from a completed transcript: each item comes with the supporting quote and its millisecond timestamp, at data.speech_understanding.response.action_items.items. The meeting-recording to task-list automation in one node. Billed per request.

AssemblyAI: LLM Chat Completion

Run a prompt through AssemblyAI's LLM Gateway (the LeMUR replacement): an OpenAI-compatible chat completion routed to Claude, GPT, Gemini and other models. Pass a transcript_id and AssemblyAI injects that transcript's text into the conversation server side, so you can ask questions about a call without templating the whole transcript through the canvas. The answer arrives at data.choices[0].message.content. Billed per token.

AssemblyAI: Speech Understanding Request

Run any speech understanding request on a completed transcript with the raw request object: speaker identification (name or role), custom formatting of dates/phone numbers/emails, or the summarization, action items and translation shapes the dedicated nodes cover. The generic escape hatch when a new understanding capability ships before its own node does. Billed per request.

AssemblyAI: Submit Transcript

Submit an audio or video file for transcription by URL. AssemblyAI downloads the file itself, so no upload step is needed: returns immediately with the new transcript's id and status queued. Poll AssemblyAI: Get Transcript with that id until status is completed (or set webhook_url and let AssemblyAI call back when done). Billed per audio duration. Reversible with AssemblyAI: Delete Transcript.

AssemblyAI: Summarize Transcript

Summarize a completed transcript with AssemblyAI's speech understanding service (the LeMUR summary replacement). Returns timestamped summary sections with headlines at data.speech_understanding.response.summarization.summary. Billed per request.

AssemblyAI: Translate Transcript

Translate a completed transcript into one or more target languages. Full translations arrive at data.translated_texts keyed by language code, with per-utterance translations alongside the original text. Billed per request.

Connecting AssemblyAI takes one credential

AssemblyAI authenticates API requests with a single API key, copied from the dashboard at assemblyai.com after signing in. Every account gets API access from signup, with $50 in free credits and no credit card required, and the same key works for both the speech-to-text API and LLM Gateway, so there is exactly one credential to manage. A detail worth knowing: AssemblyAI expects the raw key in the Authorization header with no Bearer prefix, which trips up hand-rolled integrations. Mattering formats the header for you.

1

Sign in at assemblyai.com and open your dashboard. Copy the API key shown there. New accounts get $50 in free credits at signup with no credit card required, and the key works from the moment it exists.

2

In Mattering Automations, open Connectors, click Add Connector, choose AssemblyAI, and create a credential with the API key.

3

Click Test Connection. Mattering makes a read-only call that lists your transcript jobs, which works on every account, including a brand-new one with no transcripts yet: a success proves the key, an error means it is mistyped or revoked.

4

All 14 AssemblyAI nodes appear in the canvas palette. Run AssemblyAI: Submit Transcript with a public audio URL to start your first job, and AssemblyAI: List Transcripts any time you need the ID of an existing one.

Common uses for the AssemblyAI MCP

Meeting recordings that end as task lists

A webhook carries the recording URL, AssemblyAI: Submit Transcript starts the job, and AssemblyAI: Get Transcript polls until the status is completed. Then AssemblyAI: Extract Action Items returns each follow-up with its supporting quote and millisecond timestamp, and the workflow opens a task per item in Mattering Productivity. The meeting ends, and the follow-ups already exist.

Subtitles for every published video

AssemblyAI: Get Subtitles exports a completed transcript as SRT or VTT text, ready for a video player or a publishing pipeline. The subtitle text arrives directly in the node output, capped at 200,000 characters with a truncation flag, so a downstream file step can write the .srt without any format conversion in between.

An agent that answers questions about a call

AssemblyAI: LLM Chat Completion runs a prompt through AssemblyAI's LLM Gateway, an OpenAI-compatible chat surface routed to Claude, GPT, Gemini, and other models. Pass a transcript ID and AssemblyAI injects that transcript's text into the conversation server side, so an agent can ask what the customer objected to, or what was promised and by whom, without templating the whole transcript through the canvas.

Keyword moments across every call

AssemblyAI: Word Search scans a completed transcript for words, numbers, or short phrases of up to five words, and returns each match with its occurrence count and millisecond timestamps. A scheduled sweep can run it across recent transcripts for competitor names or pricing talk, write the hits into a Mattering database, and link straight to the moment in the recording.

Recordings that are safe to share

Submit a call with PII redaction enabled and AssemblyAI beeps out the sensitive speech. AssemblyAI: Get Redacted Audio then returns the download URL for the redacted file, and because that URL expires 24 hours after it is generated, the honest pattern is to download or forward it in the same workflow run rather than storing the link.

One recording, several languages

AssemblyAI: Translate Transcript turns a completed transcript into one or more target languages in a single call, with full translations keyed by language code and per-utterance translations alongside the original text. A publishing workflow can transcribe once and produce the English notes, the Spanish summary, and the French subtitles from the same job.

Example: A meeting recording lands: transcribe it, summarize it, open the follow-ups

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

1

Trigger: an inbound webhook fires when a recording is ready, carrying the file's URL.

2

AssemblyAI: Submit Transcript sends the URL. AssemblyAI downloads the file itself, so no upload step is needed, and the node returns immediately with the new transcript's ID and a status of queued.

3

AssemblyAI: Get Transcript polls with that ID until the status reads completed. Setting a webhook URL at submit time skips the polling entirely and lets AssemblyAI call back when the job is done.

4

AssemblyAI: Summarize Transcript returns timestamped summary sections with headlines.

5

AssemblyAI: Extract Action Items returns each follow-up with its supporting quote and timestamp.

6

The workflow writes the summary into a Mattering database and opens a task per action item in Mattering Productivity, each carrying the quote that produced it.

AssemblyAI MCP questions, answered

What AssemblyAI operations are supported?

14 nodes: seven query nodes and seven action nodes. The transcript lifecycle is covered by submit, get, list, and a destructive-flagged delete. The derived reads export a completed transcript as sentences, paragraphs, or SRT and VTT subtitles, search it for words and phrases, and fetch the PII-redacted audio URL. The analysis surface runs summaries, action items, translations, chat completions with server-side transcript injection, and a generic speech understanding request for capabilities like speaker identification. Deliberately not shipped: streaming speech-to-text, raw file upload, and the retired LeMUR endpoints.

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.

AssemblyAI has an official MCP server. Why use this connector?

AssemblyAI's official MCP server gives AI tools like Claude Code a direct line to the AssemblyAI API, and for working with transcription from inside a coding environment it is a genuine option. This connector earns its place when transcription is one step in a larger workflow: the same canvas holds your other connectors and the Mattering platform itself, so a finished transcript can become tasks, database rows, and messages in the systems where the follow-through happens. Credentials live in one place with the rest of your connector secrets, and the same nodes serve visual automations and AI agents alike.

Do I need a paid AssemblyAI plan?

No. AssemblyAI gives every new account $50 in free credits at signup, with no credit card required and API access included, which is more than enough to build and test real workflows: transcription is billed per audio hour and the LLM Gateway nodes are billed per token, so a one-minute test clip costs a fraction of a cent. One free-tier limit worth knowing: free accounts process a handful of transcription jobs at a time, and extra submissions queue and process in order rather than failing.

How does an automation handle the async transcription flow?

Submitting a transcript returns immediately with the job's ID and a status of queued, and the job moves through processing to completed, or to error with the reason attached. The workflow pattern is to re-run AssemblyAI: Get Transcript with that ID until the status is completed, then branch on the result. Alternatively, pass a webhook URL at submit time and AssemblyAI calls your workflow back when the job finishes, no polling loop needed. The derived reads, like sentences, subtitles, and word search, only return data once the transcript is completed.

What happened to LeMUR?

AssemblyAI retired LeMUR on March 31, 2026, and the old endpoints no longer work. Its replacement is LLM Gateway, an OpenAI-compatible chat surface routed to Claude, GPT, Gemini, and other models, plus a speech understanding endpoint for summaries, action items, translation, and speaker identification. This connector ships the replacement surface, so workflows built on it target the current API rather than the retired one. If you are migrating an old LeMUR workflow, the chat completion node with a transcript ID covers the task and question-and-answer jobs, and the summarize node covers LeMUR summaries.

Can it transcribe live audio streams?

No. AssemblyAI's streaming speech-to-text runs over WebSocket connections, which do not fit an automation canvas built on request-and-response steps, so streaming is out of this connector's scope. The connector covers the async surface: recorded audio and video files reachable by URL, submitted with AssemblyAI: Submit Transcript and polled to completion. For recordings that arrive as files after a call or meeting ends, that is the whole job.

Is deleting a transcript reversible?

No. AssemblyAI: Delete Transcript is the connector's one destructive node: it permanently removes the transcript text, the stored audio URL, and all derived results from AssemblyAI, and the ID that remains returns placeholder values. Capture anything you need with AssemblyAI: Get Transcript first, and keep the delete step behind a condition or an approval task rather than a bare trigger. The other 13 nodes either read or perform metered, repeatable calls.

14 nodes waiting on the canvas

Put AssemblyAI 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. AssemblyAI is a product of its respective owner; Mattering is an independent platform.

Start Free Explore our Products