Speechmatics

@Speechmatics create a transcription job for this podcast episode: https://example.com/episode-42.mp3 with speaker diarization enabled

Transcription job created successfully. Job ID: jb_abc123def456. The job is now queued for processing with speaker diarization enabled. Estimated processing time is typically 15-30% of the audio duration. You can check status with the job ID or wait for completion before retrieving the transcript.

This example shows the MCP's core async workflow: submit a publicly accessible media URL, receive a job ID, then poll or wait. The API key must have write permissions. Large files may take minutes to hours depending on duration and queue depth.

@Speechmatics get the transcript for job jb_abc123def456 in plain text format and summarize the key discussion points

Transcript retrieved (3,847 words, 22:14 duration). The conversation covered three main topics: Q1 revenue projections with a focus on enterprise segment growth, upcoming product launches scheduled for May and June, and competitive positioning against two unnamed rivals. Two speakers were identified; Speaker 1 dominated with approximately 70% of speaking time. The transcript includes timestamps and confidence scores averaging 0.94.

This synthesis example pairs Speechmatics' retrieval tool with AI reasoning. The MCP returns structured JSON with speaker labels, timestamps, and confidence scores—the AI then interprets that data. Transcripts are retained for 30 days by default; retrieval after deletion returns an error.

@Speechmatics list all transcription jobs from the last 7 days and show me which ones failed or are still processing

Found 14 transcription jobs in the past 7 days. Status breakdown: 11 completed successfully, 2 currently processing (estimated completion in 4 and 12 minutes), 1 failed due to unsupported audio codec. The failed job (jb_xyz789) attempted to process a .wma file; Speechmatics requires MP3, WAV, FLAC, or M4A formats. All completed jobs have transcripts available for retrieval.

This read-heavy example demonstrates job monitoring across a project. The list tool returns newest-first with continuation cursors for pagination. Useful for auditing batch workflows or debugging stuck jobs. The MCP surfaces processing errors directly from the Speechmatics API, making codec or URL issues immediately visible.

When Speechmatics beats real-time for support call review

A 6-person support team records 40-80 customer calls per week and needs searchable transcripts for QA spot-checks and escalation handoffs. Speechmatics wins here because the async batch model lets you queue a day's calls overnight, pull structured JSON transcripts in the morning, and search them in your existing knowledge base without streaming infrastructure. The Get Usage tool makes monthly reporting trivial. The trade-off: if you need live transcription during the call for agent assist, this MCP isn't built for that—it's post-call only. But if your workflow is record-then-review, Speechmatics gives you accurate multi-language transcripts without the real-time tax. Worth it when call volume exceeds 20 per week and you're already storing recordings in a public bucket.

Why this MCP fits 2-4 person podcast teams

A three-person podcast team publishes two episodes per week and needs transcripts for show notes, SEO blog posts, and accessibility. Speechmatics handles this cleanly: drop the episode URL into Create Transcription Job, poll with Get Transcription Job until done, retrieve the SRT for captions and JSON for content extraction. The Discover Speech Features tool confirms language support before you commit to a multilingual guest series. The boundary: if you're producing daily content or need sub-15-minute turnaround, you'll hit rate limits or want a dedicated pipeline. But at 2-4 episodes per week, this MCP gives you production-grade transcripts without hiring a transcription VA. The structured JSON output integrates directly into your CMS workflow.

When batch transcription beats court reporter costs

A 10-attorney firm records 15-25 client depositions per month and needs searchable archives for case prep, but doesn't require certified court transcripts for every session. Speechmatics works when the deposition audio is already hosted (Zoom recordings in S3, for example) and you need fast, accurate text for internal review before deciding which sessions warrant a certified reporter. The Delete Transcription Job tool lets you purge sensitive transcripts after case close. The limit: this isn't a certified transcript service, so it can't replace your court reporter for trial exhibits. But for discovery-phase review and keyword search across dozens of hours of testimony, it cuts transcription costs by 70% and delivers results in hours, not days. Best fit when monthly deposition volume exceeds 10 hours.