Deepgram

This MCP server gives you access to Deepgram’s speech recognition, text-to-speech, and audio analysis capabilities through a Model Context Protocol API. It enables high-accuracy transcription, natural TTS, speaker diarization, language detection, and insightful audio analysis, all accessible from a consistent MCP client workflow.

How to use

You connect to the MCP server from an MCP client and choose the feature you want to use. Transcribe audio by sending an audio source and optional settings such as language and diarization. Generate speech from text with automatic compression to keep responses lightweight. Run advanced audio analysis to extract sentiment, topics, intents, and entities, or use speaker diarization to separate voices in a recording. You can also request model information to learn about available transcription and TTS models.

How to install

# Prerequisites
node -v
npm -v

# 1. Clone the project
git clone https://example.com/deepgram-mcp-server.git
cd deepgram-mcp-server

# 2. Install dependencies
npm install

# 3. Copy env configuration and add your API keys
cp env.example .env
# Edit .env and add:
#   DEEPGRAM_API_KEY, OPENAI_API_KEY, GROQ_API_KEY (as needed)

# 4. Build the project
npm run build

# 5. Run for HTTP transport (default port 8080)
npm start
# or
node dist/index.js

# To use a different port
node dist/index.js --port 8081

# 6. Run for STDIO transport (development)
npm run start:stdio
# or
node dist/index.js --stdio --port 8081

Additional configuration and notes

The server exposes two MCP connection methods. A remote HTTP endpoint is available at the default path provided by your deployment, and a local STDIO option lets you run the MCP process directly on your machine for development.

Security and keys

Keep your API keys secure. Store them in the environment file and limit access to trusted services. Do not commit keys to version control. Use separate keys for different environments if possible.

Usage notes about features

The server offers audio transcription, text-to-speech with automatic compression, audio analysis (including sentiment, topics, intents, and entities), speaker diarization, and language detection. It also supports multiple models optimized for various use cases and includes a smart compression system for TTS outputs to reduce transfer sizes without sacrificing perceived quality.

Troubleshooting

If you cannot reach the MCP server, confirm that the selected port is open and that the server process is running. For HTTP transport, verify the URL being used by your client. For STDIO transport, ensure the command and port you started match your client configuration.

Notes

File structure related to audio generation and compression includes separate folders for original, compressed, and decompressed audio. Review your environment paths to locate generated content after TTS requests.

Available tools

transcribe_audio

Transcribes audio to text with customizable options including language, punctuation, diarization, and sentiment analysis.

text_to_speech

Generates speech from text using Deepgram TTS models with automatic compression; outputs include paths to original and compressed audio files.

analyze_audio

Performs advanced audio analysis to extract sentiment, topics, intents, and entities from audio input.

get_models

Retrieves information about available transcription and TTS models.