Deepgram

@Deepgram transcribe this sales call recording and identify the main topics discussed: https://example.com/call-2024-01-15.mp3

I've transcribed the 18-minute call and detected four primary topics: pricing negotiation (6m 12s), product feature requests (4m 38s), implementation timeline (3m 45s), and contract terms (3m 25s). The prospect raised concerns about annual vs. monthly billing and requested custom SSO integration. Full transcript shows the conversation shifted from technical requirements to commercial terms around the 9-minute mark.

This example uses Topic Detection, which combines transcription with AI-powered topic extraction. The audio file must be publicly accessible via URL—Deepgram fetches it directly. Processing time scales with audio length; a 20-minute file typically takes 30-60 seconds. Your API key must have transcription scope enabled.

@Deepgram generate a 30-second voice message for our support team using the Aura model: 'Your ticket has been escalated to engineering. We'll follow up within 24 hours with a detailed timeline.'

I've generated the TTS audio file using Deepgram's Aura voice model. The output is a 12-second MP3 (the text was shorter than your 30-second estimate) with natural prosody and pacing suitable for customer-facing communications. The file is available at the returned URL for 24 hours. You can specify different voices (asteria, luna, stella, etc.) or adjust speaking rate if needed.

This demonstrates text-to-speech via the REST API. The response includes a temporary audio URL that expires after 24 hours—download and store it if you need persistent access. TTS requests count against your character quota. You can preview all available Aura voices by asking Deepgram to list public TTS models first.

@Deepgram show me our transcription usage for the last 30 days—I need to forecast next month's bill.

For the past 30 days across your project, you've submitted 847 hours of audio for transcription. Of that, 831 hours were successfully processed, and 824 hours are billable (some files were duplicates or below minimum duration). Your daily average is 27.5 billable hours. At your current tier rate, that projects to approximately $2,472 for a full month. Peak usage occurred on January 8th with 64 hours processed.

This uses Get Project Usage Summary to pull billing metrics. The tool returns aggregate data—you won't see per-file breakdowns here, but you get the high-level view needed for budget planning. If you manage multiple Deepgram projects, specify which project ID to query, or list all projects first to identify the right one. Usage data updates hourly, not real-time.

When Deepgram wins for support QA at scale

A 6-person support team handling 200+ customer calls weekly needs to spot escalation patterns without listening to every recording. Deepgram's topic detection and summarization tools let you batch-process call audio overnight and surface recurring issues in morning standup. The MCP exposes both transcription and topic tagging in one call, so you can filter by keyword or theme without building a pipeline. This works best when your audio is already hosted (S3, CDN) and publicly accessible—if files live behind auth, you'll spend more time wrangling URLs than analyzing calls. If your team reviews fewer than 50 calls a month, manual spot-checks are faster than setting up API keys and managing usage quotas. For teams running structured QA on hundreds of calls, Deepgram's MCP turns audio into searchable, taggable data in under an hour of setup.

Why this MCP matters for solo creators with guests

A solo podcaster recording 2-4 interviews a month wants to generate show notes and pull quotes without hiring a VA. Deepgram's summarization and text-to-speech tools let you transcribe the raw audio, extract a 3-paragraph summary, and even generate promo clips with synthetic voiceovers for social. The MCP's public models endpoint shows you which voices and transcription models are current, so you're not guessing at deprecated options. This setup shines when your episodes are under 90 minutes and you're comfortable hosting audio files on a public URL for processing. If you're editing 10+ episodes a week or need speaker diarization with custom vocabulary, you'll hit the limits of the basic toolset and want Deepgram's full API with webhooks. For the 1-4 episode-per-month creator, this MCP delivers production-ready transcripts and summaries in a single Switchy workspace without learning their entire platform.

When to use Deepgram for rep feedback loops

A 3-person sales team records discovery calls and wants the manager to review talk-time balance and objection handling without replaying 45-minute recordings. Deepgram's topic detection flags when pricing or competitor names come up, and the summarization tool gives a 4-sentence recap of each call for weekly 1-on-1s. The MCP's usage summary tool also tracks your monthly transcription spend, so you know when you're approaching your plan limit before the bill surprises you. This works when calls are stored in a shared drive with public links and your team reviews 10-30 calls a month. If you need real-time transcription during live calls or want to score sentiment on every objection, the MCP's batch-oriented tools won't keep up—you'd need Deepgram's streaming API or a dedicated conversation intelligence platform. For post-call coaching at small team scale, this MCP turns audio into coachable insights in under 5 minutes per call.