# Deepgram

Speech-to-text API that transcribes recorded or live audio into text with word timings and optional speaker labels.

- Provider: Deepgram (https://deepgram.com)
- Category: Transcribe audio
- Type: Hosted API
- Pricing model: Pay as you go from prepaid credit; Growth plan from $4,000 a year [1]
- Free tier: $200 credit that does not expire, no card [1]
- Authentication: API key in `Authorization: Token` header; temporary JWT [12]
- Connect via: REST API, WebSocket, SDKs, `dg` CLI with MCP server, docs MCP server [9]
- Read from the provider's pages on 2026-11-04
- Page: https://supermarket.dev/tools/transcribe-audio/deepgram

## Pricing

### Billing
- Pricing model: Pay as you go against prepaid credit; Growth plan from $4,000 a year in prepaid credits; Enterprise is custom [1]
- Free tier: $200 credit, does not expire (about 43,000 minutes of Nova-3) [1]
- Card required for free tier: No [1]
- Minimum spend: None on Pay as you go; Growth starts at $4,000 a year [1]
- Billing period: Per second of audio, with no block rounding for normal requests [1]

### Prices
Per hour = listed per-minute price x 60 (for example $0.0043 x 60 = $0.258). Pricing page FAQ gives $0.46 and $0.55 for streaming, rounded. The pricing page lists no price for the default `base-general` model.
| Model / mode | Pay as you go, $ per hour of audio | Growth, $ per hour of audio | Pay as you go, $ per minute |
|---|---|---|---|
| Nova-3 monolingual, pre-recorded | 0.258 | 0.216 | 0.0043 |
| Nova-3 multilingual, pre-recorded | 0.312 | 0.258 | 0.0052 |
| Whisper Large, pre-recorded | 0.288 | 0.288 | 0.0048 |
| Nova-3 monolingual, streaming | 0.462 | 0.390 | 0.0077 |
| Nova-3 multilingual, streaming | 0.552 | 0.468 | 0.0092 |
| Flux STT English, streaming | 0.462 | 0.390 | 0.0077 |
| Flux STT multilingual, streaming | 0.468 | 0.408 | 0.0078 |
| Default model `base-general` | Not published | Not published | Not published |
Source [1]

### Add-ons (streaming and pre-recorded lists are the same)
| Add-on | Pay as you go, $ per minute | Growth, $ per minute |
|---|---|---|
| Redaction | 0.0020 | 0.0017 |
| Keyterm prompting | 0.0013 | 0.0012 |
| Speaker diarization | 0.0020 | 0.0017 |
| Entity detection | 0.0017 | 0.0017 |
| Smart formatting | Included | Included |
Source [1]

## Tools

### MCP server
- Official MCP server: Yes, remote and local [9]
- Address or package: `dg mcp` (stdio, built into the `dg` CLI; the CLI also offers `dg mcp --transport sse --port 8000`); remote docs server at `https://api.dx.deepgram.com/kapa/mcp` [9]
- Remote server scope: Documentation search only; returned HTTP 401 with an OAuth challenge when sent `initialize` without credentials [21]

### MCP tools
| Tool | What it does | Read-only |
|---|---|---|
| Not listed | The `dg mcp` tool list is owned by the remote service and is not published in the docs. The docs MCP server requires OAuth, so `tools/list` could not be read without credentials. | Not stated |
Source [10]

### API
Paths for models, keys, usage and requests are from the published OpenAPI file; `/v1/listen` also serves the WebSocket.
| Endpoint | What it does |
|---|---|
| `POST https://api.deepgram.com/v1/listen` | Transcribe pre-recorded audio from a file upload or URL |
| `wss://api.deepgram.com/v1/listen` | Transcribe live audio over a WebSocket |
| `GET /v1/models` | List available models |
| `POST /v1/auth/grant` | Create a short-lived JWT (30-second TTL) |
| `GET /v1/auth/token` | Check that an API key is valid |
| `/v1/projects/{project_id}/keys` | Manage API keys |
| `/v1/projects/{project_id}/usage` | Read usage |
| `/v1/projects/{project_id}/requests` | List past requests (logs kept 90 days) |
Source [4]

### Features
- Default model: `base-general` [4]
- Default model (other page): `base` when the `model` parameter is omitted [3]
- Languages: Nova-3 supports 60+ languages, including multilingual real-time transcription with `language=multi`; Flux multilingual covers 10 languages [1]
- Speaker labels: Yes, via `diarize_model` (batch and streaming); each word gets a speaker number starting at 0 [4]
- Word timestamps: Yes, each word has `start` and `end` in seconds and a confidence value [3]
- Live streaming: Yes, WebSocket at `/v1/listen` with Nova-3 and Flux STT [5]
- Max file size: 2 GB for pre-recorded requests; requests over 10 minutes of processing (Nova, Base, Enhanced) or 20 minutes (Whisper) return `504` [3]
- Audio input: MP3, MP4, MP2, AAC, WAV, FLAC, PCM, M4A, Ogg, Opus, WebM; over 100 formats and encodings; file upload or URL [16]

## Authentication

### Methods
| Method | Used for | How |
|---|---|---|
| API key | REST and WebSocket calls | Header `Authorization: Token YOUR_DEEPGRAM_API_KEY`; HTTPS only |
| Temporary JWT | Client-side use with short-lived access | Create with `POST /v1/auth/grant` using an API key; send `Authorization: Bearer <JWT>`; 30-second TTL |
| Stored CLI credentials | `dg mcp` local MCP server | `dg login`, or `DEEPGRAM_API_KEY` in CI |
| OAuth | Remote docs MCP server | Server answers `401` with a `WWW-Authenticate: Bearer resource_metadata` challenge |
Source [12]

### Getting access
- Who creates the account: A person signs up in the Console; whether an agent can sign up alone is not stated [14]
- Where to get a key: Deepgram Console (`console.deepgram.com`) or the Deepgram API [12]
- Other requirements: None stated for the free credit; no credit card needed to start [1]

## Setup

### Addresses
- REST API: `https://api.deepgram.com/v1/listen`; regional hosts `api.eu.deepgram.com`, `api.au.deepgram.com`, `api.in.deepgram.com` [2]
- MCP server: `dg mcp` (local, stdio); `https://api.dx.deepgram.com/kapa/mcp` (remote, docs only) [9]
- OpenAPI spec: `https://developers.deepgram.com/openapi.json` [18]
- Docs for agents: `https://developers.deepgram.com/llms.txt`; append `.md` to any docs page for Markdown [18]
- Status page: `https://status.deepgram.com` [21]

### Install
| What | Command |
|---|---|
| MCP for Claude Code (docs server) | `claude mcp add deepgram-docs --scope project --transport http https://api.dx.deepgram.com/kapa/mcp` |
| CLI with MCP server, macOS or Linux | `curl -fsSL https://deepgram.com/install.sh \| sh` |
| CLI via pip | `pip install deepctl` |
| CLI via Homebrew | `brew tap deepgram/tap` then `brew install deepgram` |
| Agent skills for Claude Code | `/plugin marketplace add deepgram/skills` then `/plugin install deepgram@deepgram-agent-skills` |
| JavaScript SDK | `npm install @deepgram/sdk` |
| Python SDK | `pip install deepgram-sdk` |
| .NET SDK | `dotnet add package Deepgram` |
| Go SDK | `go get github.com/deepgram/deepgram-go-sdk` |
Source [9]

## Limits

### Rate limits
- Concurrent pre-recorded requests: 50 per project for Nova-3, Nova-2, Nova, Enhanced and Base (Pay as you go and Growth) [2]
- Concurrent pre-recorded requests (other page): 100 per project for Nova, Base and Enhanced [3]
- Concurrent streaming connections: 150 on Pay as you go, 225 on Growth (North America); Enterprise set to peak traffic [1]
- Concurrent requests with diarization: 50 pre-recorded and 50 streaming in North America; 25 each in Europe, Australia and India [2]
- Concurrent Whisper Cloud requests: 3 pre-recorded; not available outside North America [2]
- Scope of limits: Per project, not per account or API key; extra projects on self-serve accounts get 1 concurrent stream [2]

### Other limits
- Pre-recorded file size: 2 GB [3]
- Pre-recorded processing time: Requests over 10 minutes (Nova, Base, Enhanced) or 20 minutes (Whisper) return `504 Gateway Timeout` [3]
- Temporary token lifetime: 30 seconds [13]
- Log retention: Request metadata and usage logs retrievable for 90 days [6]

## Privacy

### Your data
- Data location: Global endpoint `api.deepgram.com` by default; regional endpoints for EU, Australia and India keep processing and storage in-region [6]
- Retention: By default audio and transcripts are retained to improve models; with `mip_opt_out=true` only for the time needed to process the request. Usage logs: 90 days [6]
- Used for training: Yes by default through the Model Improvement Program; opt out per request with `mip_opt_out=true` [6]
- Zero data retention: Available per request with `mip_opt_out=true`; account-wide enforcement by contacting Deepgram [6]

### Compliance
- Certifications: SOC 2 Type 1 and Type 2; GDPR ready; CCPA compliant; PCI compliant [7]
- HIPAA: Business Associate Agreement available on request [7]
- Encryption: TLS 1.3 in transit and AES-256 at rest [6]

## Sources

1. Pricing: https://deepgram.com/pricing
2. API rate limits: https://developers.deepgram.com/reference/api-rate-limits
3. Pre-recorded audio: https://developers.deepgram.com/docs/pre-recorded-audio
4. Listen (pre-recorded) reference: https://developers.deepgram.com/reference/speech-to-text/listen-pre-recorded
5. Listen (live audio) reference: https://developers.deepgram.com/reference/speech-to-text/listen-streaming
6. Your data at Deepgram: https://developers.deepgram.com/trust-security/your-data
7. Data privacy compliance: https://developers.deepgram.com/trust-security/data-privacy-compliance
8. Model Improvement Partnership Program: https://developers.deepgram.com/docs/the-deepgram-model-improvement-partnership-program
9. Agentic developer tools: https://developers.deepgram.com/developer-tools/agentic-tools
10. CLI MCP server: https://developers.deepgram.com/developer-tools/cli/mcp-server
11. CLI installation: https://developers.deepgram.com/developer-tools/cli/installation
12. Authenticating: https://developers.deepgram.com/guides/fundamentals/authenticating
13. Token-based auth: https://developers.deepgram.com/guides/fundamentals/token-based-authentication
14. Make your first API request: https://developers.deepgram.com/guides/fundamentals/make-your-first-api-request
15. Models and languages overview: https://developers.deepgram.com/docs/models-languages-overview
16. Supported audio formats: https://developers.deepgram.com/docs/supported-audio-formats
17. Regional endpoints: https://developers.deepgram.com/reference/regional-endpoints
18. llms.txt: https://developers.deepgram.com/llms.txt
19. OpenAPI JSON: https://developers.deepgram.com/openapi.json
20. Status page: https://status.deepgram.com
21. Checked by supermarket.dev