AssemblyAI
AssemblyAI provides speech-to-text transcription and audio intelligence APIs for uploaded and remotely hosted media.
Verdict
Common use cases
- Summarize hour-long sales calls in seconds
- Search customer support recordings for feature requests
- Generate SRT subtitles for training videos
- Redact PII from compliance call recordings
- Extract action items from weekly standups
Integration
- Vendor
- AssemblyAI
- Category
- other
- Auth
- API_KEY
- Tools
- 9
- Composio slug
assemblyai
Tools
- Delete Transcript Contentdestructive
Irreversibly delete a transcript's stored content. AssemblyAI retains a tombstone that later GET and list calls can still return, and those later responses may omit the is_deleted marker.
- Export Subtitles
Export a completed transcript as raw SRT or WebVTT subtitle text.
- Get Redacted Audio
Get the processing status and short-lived signed download URL for PII-redacted audio. The transcript must have been submitted with text and audio PII redaction enabled and at least one explicit redaction policy. Treat the returned URL as tr
- Get Transcript
Retrieve an AssemblyAI transcript's current state and results, optionally waiting until asynchronous processing completes or fails.
- Get Transcript Segments
Return sentence- or paragraph-level segments, timestamps, confidence, and words for a completed transcript.
- List Transcripts
List project transcripts from the last 90 days, newest first, with optional filters and a cursor for the next older page.
- Search Transcript
Find words or phrases in a completed transcript and return occurrence counts and timestamps.
- Submit Transcript
Queue an asynchronous pre-recorded transcription from a reachable media URL. Returns immediately with the transcript ID and processing status; use GET_TRANSCRIPT to wait for results.
- Upload Media
Upload an audio or video file to AssemblyAI and return the project-private URL to use when submitting a transcript. AssemblyAI provides no API to delete this uploaded object; it remains subject to provider retention.
Setup
Setup guide
- 11. Sign in to your AssemblyAI account at assemblyai.com and navigate to the API Keys section of your dashboard to copy your key. 2. In Switchy, open Workspace Settings and click Integrations, then find AssemblyAI in the list and click Connect. 3. Paste your API key into the prompt and click Authorize — Switchy validates the key by listing recent transcripts. 4. Open any Project and type '@AssemblyAI submit a transcript from https://example.com/meeting.mp3' to queue your first transcription job; the MCP returns a transcript ID. 5. Wait a few minutes, then type '@AssemblyAI get transcript <ID>' to retrieve the completed text and metadata. 6. Test search by typing '@AssemblyAI search transcript <ID> for "budget"' to find every mention with timestamps.
What teammates see: by default, memories from AssemblyAI are scoped to the Project (PROJECT visibility) - you can mark any memory PRIVATE or share it ORG-wide.
Works well with
Top models
Compatibility data appears once enough Projects have used this MCP together with a given model.
How Switchy teams use it
Starter prompts
Transcribe Meeting Recording
@AssemblyAI submit a transcript from https://example.com/team-sync.mp4 with speaker diarization enabled, then wait for completion and show me the full text organized by speaker.
Find Customer Pain Points
@AssemblyAI list the five most recent transcripts, then search each one for the phrases 'frustrated', 'doesn't work', and 'bug' — give me a summary of which calls mentioned problems and at what timestamps.
Export Video Subtitles
@AssemblyAI get transcript <ID> and export it as WebVTT subtitles, then show me the first twenty lines so I can verify timing accuracy.
Redact Sensitive Information
@AssemblyAI submit a transcript from https://example.com/support-call.wav with PII redaction for names, phone numbers, and credit cards enabled for both text and audio, then retrieve the redacted audio download URL.
Segment Long Podcast
@AssemblyAI get transcript <ID> and return paragraph-level segments with start and end times — I want to identify the three longest segments for potential highlight reels.
Example outputs
Illustrative - representative of the model's voice and quality, not literal recordings.
@AssemblyAI transcribe the customer call recording at https://example.com/calls/2024-03-15.mp3 with speaker labels and detect action items
Transcript queued with ID abc123xyz. Processing typically completes in 15–30% of the audio duration. Once ready, the transcript will include speaker diarization (Speaker A, Speaker B, etc.) and any detected action items flagged by the LeMUR model. You can poll the status or wait for completion; I'll retrieve the full results when processing finishes.
This example shows the async workflow: submit returns immediately with a transcript ID, then you must poll or wait. The MCP requires a publicly reachable media URL—local files need pre-upload to S3 or similar. Speaker labels and action-item detection are opt-in features that may incur additional processing time and cost.
@AssemblyAI search all transcripts from the last 30 days for mentions of 'refund policy' and show me the top three with timestamps
Found 'refund policy' in 7 transcripts. Top three by relevance: (1) Transcript def456 from March 10—two mentions at 02:14 and 08:47. (2) Transcript ghi789 from March 8—one mention at 05:32. (3) Transcript jkl012 from March 3—three mentions at 01:05, 03:22, and 09:10. Each timestamp links to the sentence-level segment where the phrase appears.
This combines list and search tools to surface historical audio insights. AssemblyAI retains transcripts for 90 days; older recordings won't appear. Search is case-insensitive and matches whole or partial phrases. Rate limits apply when iterating large transcript sets—batch requests if you expect hundreds of results.
@AssemblyAI export the last board meeting transcript as SRT subtitles and summarize the key decisions in three bullet points
SRT file ready (transcript mno345, 47 minutes, 1,203 subtitle cues). Key decisions: • Approved Q2 marketing budget increase of 18% to accelerate EU expansion. • Deferred the vendor consolidation proposal until legal review completes in April. • Established a cross-functional task force to evaluate the new CRM migration timeline. The subtitle export preserves speaker labels and timestamps; you can import it directly into video editing tools.
This pairs a data-export tool (subtitles) with the AI's reasoning layer to distill actionable insights. The transcript must be fully processed before export succeeds. SRT and WebVTT are the only supported formats. If the original transcription didn't enable speaker diarization, the subtitle cues won't include speaker tags.
Use-case deep-dives
When AssemblyAI beats manual call review for support teams
A 6-person support team records 40-60 customer calls per week and needs to spot recurring objections or feature requests without listening to every call. AssemblyAI wins here because you submit the call URL, wait for the transcript, then use the search tool to find every mention of "pricing" or "integration" across the entire batch in under a minute. The redacted audio tool strips PII before you share clips with product. The threshold: if your team is under 20 calls per week, manual spot-checking is faster than wiring up the API. Above that volume, the search and segment tools pay for themselves by surfacing patterns you'd miss in a spreadsheet of notes. Set up a shared Switchy workspace where anyone can paste a call URL and get the searchable transcript without touching code.
How a two-person podcast uses AssemblyAI for show notes
A weekly podcast with one host and one editor ships episodes to YouTube, Spotify, and a blog. The editor pastes the raw audio URL into Switchy, submits the transcript, then exports SRT subtitles for the YouTube upload and pulls paragraph segments for the blog show notes—all in one 10-minute session while the host reviews the final cut. AssemblyAI handles speaker labels and timestamps automatically, so the editor isn't retyping or guessing who said what. The trade-off: if your podcast is under 20 minutes and you already have a tight manual workflow, the setup overhead isn't worth it. Past 30 minutes per episode or two episodes per week, the segment and subtitle export tools cut production time in half. Use Switchy to keep the transcript, subtitles, and show-note draft in one thread so the host can review without switching tools.
When AssemblyAI replaces post-meeting note reconciliation
A 5-person remote product team runs three standups per week and used to spend 15 minutes after each call reconciling notes in Notion. Now they record the Zoom call, drop the MP4 URL into Switchy, and let AssemblyAI transcribe while they move to the next meeting. The get-segments tool returns paragraph-level chunks with timestamps, so the PM copies the relevant decisions into Linear without re-listening. The search tool finds every mention of a ticket number or feature name across the last month of standups when someone asks "didn't we already discuss this?" The boundary: if your meetings are under 15 minutes or you have a dedicated note-taker, manual notes are cleaner. Above 20 minutes or when the team is spread across time zones and can't all attend live, the transcript becomes the single source of truth. Keep the transcripts in a Switchy workspace so the whole team can search old standups without asking the PM.
Frequently asked
What does the AssemblyAI MCP do in Switchy?
It lets your team transcribe audio and video files, search transcripts for specific words, export subtitles, and redact PII from both text and audio. You submit a media URL, AssemblyAI processes it asynchronously, then you retrieve the transcript with timestamps and confidence scores. Useful for meeting recordings, podcast production, or compliance workflows where you need searchable text from audio.
Do I need a paid AssemblyAI account to use this MCP?
Yes. You need an AssemblyAI API key, which requires a paid account after the free tier runs out. The MCP authenticates with that key, so whoever connects it must have access to your organization's AssemblyAI dashboard. Free-tier limits apply to transcription minutes and concurrent jobs, not to Switchy itself.
Can it transcribe live audio or only pre-recorded files?
Only pre-recorded files. You submit a publicly reachable media URL and AssemblyAI processes it in the background. If you need real-time streaming transcription, use AssemblyAI's WebSocket API directly—this MCP only wraps the asynchronous batch endpoint. Good for post-call analysis, not for live captions during a Zoom call.
How is this better than uploading audio to AssemblyAI's dashboard?
It keeps transcription in your team's AI workflow. Instead of switching tabs to paste URLs and copy results, you submit jobs and retrieve transcripts inside Switchy prompts. The MCP also exposes search and segment tools, so you can ask the AI to find specific phrases or export subtitles without leaving the conversation.
Who on the team should connect the AssemblyAI MCP?
Whoever owns your AssemblyAI API key and understands your transcription budget. That person's key will be used for all team requests, so make sure they're comfortable with usage showing up under their account. Transcription costs are billed by AssemblyAI, not Switchy, so coordinate with finance if you expect high volume.