How to Connect Deepgram to ChatGPT via MCP

💡 No install? Use cloud.anythingmcp.com directly. Sign in, click Connectors → Deepgram, paste your credentials, mint an MCP API key and you're done. No Docker, no git clone, no local server.

Deepgram + ChatGPT

Deepgram speech AI with your own API key: transcribe recordings from a URL with speaker labels, summaries, topics and sentiment, analyze text, turn text into speech with Aura voices, and check models, keys, members, usage, requests, spend and credit balances.

Prerequisites

See the full setup instructions baked into the connector (visible in the in-app store when you select the connector). The required environment variables for this connector are:

DEEPGRAM_API_KEY

Step 1 — Get credentials

Setting up takes about 3 minutes. You need a Deepgram account; a new account starts with free credit.

1. Sign in at https://console.deepgram.com and open the project you want to use.
2. Open API Keys, click Create a New API Key, give it a name and pick the role. Member is enough for transcription, text to speech, text intelligence, keys and usage; Admin or Owner is needed for members, balances, spend and purchases (billing:read). Leave the expiration empty or set one you will remember.
3. Copy the key, which is shown only once, into DEEPGRAM_API_KEY. Paste the key itself: the connector sends it as Authorization: Token <key>.
4. Install the connector and run deepgram_list_projects to check the key.

What it costs: deepgram_transcribe_url is billed per audio minute (more with summaries and other intelligence features), deepgram_analyze_text per token and deepgram_text_to_speech per character, all to the project of the key. Every other tool is free. For a long recording or a large batch the AI says roughly what it will cost before running it, and deepgram_list_balances shows the credit left.

What it does:

• Transcribe a recording with deepgram_transcribe_url from a public url; a presigned S3, GCS or Drive link works, a page behind a sign-in does not. Good defaults are model: nova-3 and smart_format: true. diarize with utterances or paragraphs shows who said what, detect_language handles an unknown language (or language: multi on nova-3 for mixed languages), keyterm helps with names and jargon, summarize: v2, topics, intents and sentiment analyze calls (English only), and redact hides card numbers or personal data.
• Analyze text you already have, such as a transcript, an email or notes, with deepgram_analyze_text and summarize, topics, intents or sentiment (English).
• Make audio with deepgram_text_to_speech and an Aura voice in model; deepgram_list_models lists the voices with language, accent and a sample link.
• Models: deepgram_list_models and deepgram_get_model for the public models, deepgram_list_project_models for what one project can use, custom models included.
• Account: deepgram_list_projects for the project_id, then the project, its API keys (read only, the secret is never returned) and its members.
• Usage and cost: deepgram_get_usage_breakdown for hours, requests, tokens and characters per period or grouped by endpoint, model, key or tag; deepgram_list_usage_fields for the values to filter on; deepgram_list_requests and deepgram_get_request for single calls with their cost; deepgram_get_billing_breakdown for spend in USD; deepgram_list_balances, deepgram_get_balance and deepgram_list_purchases for credit.

Every tool is on at install. The three billed tools spend the project's credit, so the AI confirms large or repeated runs with you first.

Data formats: the transcript is in results.channels[0].alternatives[0].transcript with a confidence from 0 to 1, one channel per audio channel with multichannel. With paragraphs the formatted text with speaker labels is in alternatives[0].paragraphs.transcript; results.utterances lists turns with start and end in seconds and speaker (0, 1, ...). Summaries are in results.summary.short, topics, intents and sentiment in results.topics, results.intents and results.sentiments. Per-word timings are removed to keep answers small, so the AI uses utterances or paragraphs for timing. metadata.duration is the audio length in seconds and metadata.request_id identifies the call in the request log. For text intelligence the summary is in results.summary.text. Text to speech returns the audio as a file attached to the result (MP3 by default, up to 8 MiB); Deepgram does not store it, so keep the file. Usage hours come as decimal strings, billing amounts and balances in units (usually usd), dates as YYYY-MM-DD in UTC.

Paging: deepgram_list_requests pages with page (from 0) and limit (up to 1000); the AI keeps going while a page comes back full. The other lists return everything at once.

Limits: text to speech takes at most 2,000 characters per call, so the AI splits longer text at sentence boundaries. Transcription answers in one call and the connector waits up to 30 seconds: nova-3 usually transcribes an hour of audio well within that, but very long files with many intelligence features can take longer and are better transcribed in parts. Pay-as-you-go and Growth projects run up to 50 concurrent transcriptions, 15 text-to-speech and about 10 text intelligence requests at once. Text intelligence takes up to 150K tokens of input.

Common errors: 400 with err_code and err_msg means an option the model does not support (for example keyterm on nova-2, or intelligence features on non-English audio) or a URL Deepgram cannot download; 400 TOKEN_LIMIT_EXCEEDED means text over 150K tokens. 401 is a wrong or expired key, 402 a project without credit, 403 a key whose role lacks the scope (members and billing need Admin or Owner), 404 an unknown project, request or balance id, 413 text over 2,000 characters for speech, 429 too many requests at once.

The connector does not stream live audio, does not use the voice agent API, does not upload local files (only URLs) and does not create or delete keys or members.

Step 2 — Install the adapter

curl -fsSL https://raw.githubusercontent.com/HelpCode-ai/anythingmcp/main/docker-compose.quickstart.yml -o docker-compose.yml
printf 'JWT_SECRET=%s\nENCRYPTION_KEY=%s\n' "$(openssl rand -hex 32)" "$(openssl rand -hex 32)" > .env
docker compose up -d

Step 3 — Add the connector in ChatGPT

Your server URL: in AnythingMCP, open MCP Servers → the server this connector is on, and copy its URL (https://cloud.anythingmcp.com/mcp/…). Use it wherever this guide shows YOUR_SERVER_ID.

ChatGPT adds MCP servers as plugins on its Plugins page, on the web and on paid plans. Developer mode is no longer needed. In a Business workspace only admins and owners can add one; in Enterprise and Edu an admin can grant the Create plugins with MCPs permission.

1. Go to chatgpt.com/plugins, click Add, then Add custom MCP server.
2. Enter a name, e.g. Name: Deepgram.
3. Under Connection, enter the Server URL https://cloud.anythingmcp.com/mcp/YOUR_SERVER_ID and keep Authentication on OAuth.
4. Select I understand and want to continue, then Create as a plugin. When ChatGPT asks, sign in to AnythingMCP and allow access.
5. Install the new plugin from your personal plugins, then type @ in a chat and pick it.

Available tools

Example prompts

• "Transcribe the call at https://example.com/calls/2026-10-29.mp3 with speaker labels and give me a short summary."
• "Which topics and what sentiment come up in this support transcript? Here is the text."
• "Read this paragraph aloud with an English Aura voice and give me the MP3."
• "How many hours of audio did our project transcribe this month, grouped by model?"
• "How much credit is left on the project, and what did we spend last week in USD?"

FAQ

Does the free ChatGPT plan work with custom MCP servers? OpenAI documents custom MCP servers for paid plans only; the Free plan is not on its list.

Next steps

• Connector landing page
• Connect to Claude
• Connect to Cursor
• Connect to Gemini
• Connect to Meta Muse
• Connect to Microsoft Copilot