MCP server
Deepgram
Cloud transcription inside an AI client, suitable only for approved non-sensitive audio.
Published by the company that makes the product.
What it can access
Uses a Deepgram API key and sends the selected audio to Deepgram. The AI client receives the transcript.
Security note
This is cloud transcription. Do not use it for testimony or protected interviews without an approved contract, consent, retention review, and redaction plan. Prefer local Whisper for the sensitive lane.
Do you need it?
Connect only when approved cloud transcription volume justifies it. For protected interviews, use local Whisper instead.
Capability and permissions
The server uses a Deepgram API key and sends selected audio to Deepgram. The transcript returns to the AI client, so both services participate in the data path.
Least-privilege setup
As of 2026-09-01, install only from Deepgram’s official repository. Create a separate low-limit API key, store it in the client’s secret field rather than a document, and expose only a folder of public test audio.
Harmless success test
Record 20 seconds of invented, consented speech. Transcribe it and compare every word. Confirm no other audio file was read.
Disable, revoke, remove
Remove the MCP configuration, restart the client, revoke the Deepgram API key, and delete the test audio and transcript if no longer needed.
When not to connect
Do not connect for testimony, source interviews, protected meetings, health information, or audio whose speakers did not consent to cloud processing.