# Deepgram MCP for AI Agents MCP > Deepgram gives your AI agents full control over high-speed audio processing. It lets you transcribe remote audio streams (WAV, MP3) using the Nova-2 model and generate professional speech from raw text using Aura voices. Beyond transcription and synthesis, this MCP handles core account functions: managing API keys, tracking usage across projects, monitoring credit balances, and inviting team members. ## Overview - **Category:** ai-frontier - **Price:** Free - **Tags:** speech-to-text, text-to-speech, transcription, voice-ai, natural-language-processing, audio-processing ## Description Deepgram lets your AI agent handle complex audio workflows right inside your chat environment. You don't have to leave your development tool or dashboard just because you need to transcribe a recording or synthesize voiceover content. Instead of manually uploading files to an external portal and waiting for batch processing, your agent sends the request directly. It can pull transcriptions from remote URLs—supporting formats like WAV or MP3—using their fast Nova-2 model. For speech generation, it converts plain text into high-fidelity audio streams using Aura voices. And since building these systems means managing access and costs, this MCP also lets you manage the underlying infrastructure. You can list project API keys, create new ones with specific scopes, or check current credit balances to ensure your pipelines never drop due to limits. Because Vinkius hosts and manages these connections, it brings all these critical audio AI functions—from transcription to key management—into one place for your agent to access. ## Tools ### send_invite Sends an invitation to add a team member to a Deepgram project. ### list_keys Retrieves a list of all active API keys for the current project. ### get_balances Checks and reports the remaining credit balance for a specific Deepgram project. ### create_key Generates a new API key, allowing you to specify its comment, required scopes, or an optional expiration date. ### delete_key Permanently revokes and deletes an existing API key for security purposes. ### list_members Shows all current team members who have access to the project. ### list_projects Provides a comprehensive list of all Deepgram projects associated with your account. ### transcribe_url Transcribes audio content from any accessible URL, supporting formats like WAV, MP3, and FLAC using the Nova-2 model. ### speak_text Converts provided text into high-fidelity speech audio data using Deepgram's Aura voices. ### get_usage Retrieves detailed API usage statistics for a project, allowing filtering by date ranges. ## Prompt Examples **Prompt:** ``` Can you transcribe this podcast audio and tell me what the main points were? ``` **Response:** ``` **Transcription Report (Nova-2 Model)** * **Source:** https://podcast.com/ep14 * **Time Processed:** 35 minutes, 12 seconds * **Key Takeaways:** * AI is moving past chatbots into agentic workflows. * The best data pipelines require constant usage monitoring. * Focus on key management to secure your infrastructure. (Confidence Score: 94%) ``` **Prompt:** ``` I need a high-fidelity voiceover for this paragraph about our new product. ``` **Response:** ``` **Audio Generation Successful!** * **Text Input:** "The future of AI is agentic and requires secure, scalable audio processing." * **Voice Used:** Aura Asteria (English) * **Output File:** `audio_v2.mp3` (Binary stream provided for download). * **Notes:** The synthesized audio maintains a natural pace and high fidelity, ready for immediate use in your application's UI. ``` **Prompt:** ``` What was my deepgram usage last month and do I have enough credit? ``` **Response:** ``` Checking Deepgram usage for the period 2026-05-01 to 2026-05-31... **Usage Summary:** * Total Transcription Time: 480 minutes (up from 390 last month). * Total TTS Generation Bytes: 18 MB. * Current Balance: **$45.50 credits remaining.** Looks good; your current spending keeps you well within the project's overall limits. ``` ## Capabilities ### Transcribe Audio from URLs Send automated requests to transcribe audio files hosted at a specific URL using the Nova-2 model. ### Generate Speech from Text Convert raw text into high-quality, natural-sounding speech audio streams using Deepgram's Aura voices. ### Monitor API Usage and Limits Analyze detailed usage statistics for a project, mapping transcription time and TTS byte consumption over custom date ranges. ### Manage Project Access Keys List existing Deepgram access keys or create entirely new ones with specific scopes and expiration dates. ### Manage Team Membership View all team members associated with a project or send invitations to expand the development team's access. ## Use Cases ### Analyzing Customer Feedback Recordings A product manager needs to analyze 50 hours of customer call recordings. Instead of manually processing each file, they ask their agent to use `transcribe_url` on a batch of recording links. The agent returns structured text, allowing the PM to instantly categorize pain points. ### Building Voice-Activated Tutorials A developer needs to create an internal training module with custom voice narration. They feed marketing copy into `speak_text` and generate MP3 assets on demand, which are then integrated into the application's help flow. ### Auditing Multi-Tenant Billing An operations team needs to check if a specific department exceeded its allocated audio budget last quarter. They use `get_usage` with precise date filters to map out exactly where the consumption occurred, preventing unexpected overages. ### Onboarding New Development Staff A lead engineer needs to grant a new team member access to the main audio project. They use `list_members` to confirm who is on board and then execute `send_invite` to securely onboard the newcomer. ## Benefits - Get instant transcriptions: Instead of uploading files, you simply provide a URL, and the agent uses `transcribe_url` to get text from audio streams. - Save time on content creation: Use `speak_text` to convert any block of copy into natural-sounding speech audio instantly for voiceovers or alerts. - Maintain security and control: You can use `list_keys`, `create_key`, and `delete_key` to manage access credentials directly within your workflow. - Keep costs under wraps: Check project limits and credit balances using `get_balances` and `get_usage` before running expensive pipelines. - Scale development teams: Use the team management tools, like listing members or sending invites via `send_invite`, so multiple people can work on the same audio projects. ## How It Works The bottom line is that you manage complex, multi-step audio AI tasks and billing oversight entirely through conversation. 1. Subscribe to this MCP and input your Deepgram API Key, which you find in the Deepgram Console under Settings > API Keys. 2. Your AI client uses natural conversation to determine if it needs to transcribe audio from a URL or generate speech from text. It then executes the required tool calls against Deepgram's services. 3. The agent returns the requested output—either structured transcription data, binary audio streams, or usage reports—directly in your chat interface. ## Frequently Asked Questions **How can I use the Deepgram MCP to transcribe audio from a website link?** You just tell your agent the URL and the request. It handles connecting to that remote stream, transcribing the content using Nova-2, and returning clean text data directly into your chat window. **I need synthetic voices for my app; what is the easiest way with Deepgram MCP?** Simply provide the text you want spoken. The agent uses the `speak_text` tool to generate high-quality MP3 binary data using Aura voices, which you can then embed directly into your application. **How do I manage billing and API keys when using Deepgram MCP?** The MCP gives you full oversight. You can run `get_usage` to see exactly how much audio was processed over a date range, or use `create_key` to generate new access credentials while keeping the old ones secure. **Does Deepgram MCP help me manage who on my team has access?** Yes. You can list all current project members using `list_members`, and if you need to add someone, you just send an invitation using the `send_invite` tool. **What if I run out of Deepgram credits? Can the MCP help me track that?** Absolutely. You can use `get_balances` to check your current credit status and `get_usage` to see spending patterns, so you always know when you'll need a top-up.