SIGN UPLOG IN
X (Twitter)LinkedInGitHubDiscordYouTube

How to integrate Deepgram MCP with Claude Agent SDK

Connect Claude Agent SDK to Deepgram MCP. Transcribe meeting audio from this url, summarize podcast episode for quick review, and more using natural language, with authentication handled for you.

Get started for freeGet a demo
Deepgram logoDeepgram
Api Key

Deepgram is an AI-powered speech recognition platform for accurate audio transcription and understanding. It enables fast, scalable speech-to-text with advanced audio intelligence features.

17 Tools

Introduction

This guide walks you through connecting Deepgram to the Claude Agent SDK using the Composio tool router. By the end, you'll have a working Deepgram agent that can transcribe audio, convert text to speech, browse available models, and track project usage and billing through natural language commands.

This guide will help you understand how to give your Claude Agent SDK agent real control over a Deepgram account through Composio's Deepgram MCP server.

Before we dive in, let's take a quick look at the key ideas and tools involved.

Also integrate Deepgram with

Claude CodeClaude CoworkOpenAI Agents SDKCodexOpenClawHermesAtomic AgentGoogle ADKLangChainVercel AI SDKMastra AILlamaIndexCrewAIPydantic AIAutoGen

TL;DR

Here's what you'll learn:
  • Get and set up your Claude/Anthropic and Composio API keys
  • Install the necessary dependencies
  • Initialize Composio and create a Tool Router session for Deepgram
  • Configure an AI agent that can use Deepgram as a tool
  • Run a live chat session where you can ask the agent to perform Deepgram operations

What is Claude Agent SDK?

The Claude Agent SDK is Anthropic's official framework for building AI agents powered by Claude. It provides a streamlined interface for creating agents with MCP tool support and conversation management.

Key features include:

  • Native MCP Support: Built-in support for Model Context Protocol servers
  • Permission Modes: Control tool execution permissions
  • Streaming Responses: Real-time response streaming for interactive applications
  • Context Manager: Clean async context management for sessions

What is the Deepgram MCP server, and what's possible with it?

The Deepgram MCP server is an implementation of the Model Context Protocol that connects your AI agent and assistants like Claude, Cursor, etc directly to your Deepgram account. It provides structured and secure access to Deepgram's speech AI services, so your agent can transcribe audio, generate summaries, convert text to speech, detect topics, and analyze project usage on your behalf.

  • Automated audio transcription: Your agent can transcribe pre-recorded audio files into accurate, readable text for effortless documentation and analysis.
  • Audio summarization: Have your agent generate quick, concise summaries of audio content, making it easy to review lengthy recordings in seconds.
  • Natural-sounding text-to-speech: Let your agent convert written text into lifelike speech audio using Deepgram's public TTS models for accessibility or creative applications.
  • Topic detection from audio: Enable your agent to identify and report on the main topics discussed within audio files, streamlining content review and categorization.
  • Usage analytics and project insights: Retrieve and analyze project usage metrics, scope permissions, and available models to optimize your Deepgram integration and monitor account health.

What is the Composio tool router, and how does it fit here?

What is Composio SDK?

Composio's Composio SDK helps agents find the right tools for a task at runtime. You can plug in multiple toolkits (like Gmail, HubSpot, and GitHub), and the agent will identify the relevant app and action to complete multi-step workflows. This can reduce token usage and improve the reliability of tool calls. Read more here: Getting started with Composio SDK

The tool router generates a secure MCP URL that your agents can access to perform actions.

How the Composio SDK works

The Composio SDK follows a three-phase workflow:

  1. Discovery: Searches for tools matching your task and returns relevant toolkits with their details.
  2. Authentication: Checks for active connections. If missing, creates an auth config and returns a connection URL via Auth Link.
  3. Execution: Executes the action using the authenticated connection.

Step-by-step Guide

Step by step09 STEPS
1

Prerequisites

Before starting, make sure you have:
  • Composio API Key and Claude/Anthropic API Key
  • Primary know-how of Claude Agents SDK
  • A Deepgram account
  • Some knowledge of Python
2

Getting API Keys for Claude/Anthropic and Composio

Claude/Anthropic API Key
  • Go to the Anthropic Console and create an API key. You'll need credits to use the models.
  • Keep the API key safe.
Composio API Key
  • Log in to the Composio dashboard.
  • Navigate to your API settings and generate a new API key.
  • Store this key securely as you'll need it for authentication.
3

Install dependencies

npm install @anthropic-ai/claude-agent-sdk @composio/core dotenv

Install the Composio SDK and the Claude Agents SDK.

What's happening:

  • @composio/core provides Composio integration for Anthropic
  • @anthropic-ai/claude-agent-sdk is the core agent framework
  • dotenv/config loads environment variables
4

Set up environment variables

bash
COMPOSIO_API_KEY=your_composio_api_key_here
USER_ID=your_user_id_here
ANTHROPIC_API_KEY=your_anthropic_api_key_here

Create a .env file in your project root.

What's happening:

  • COMPOSIO_API_KEY authenticates with Composio
  • USER_ID identifies the user for session management
  • ANTHROPIC_API_KEY authenticates with Anthropic/Claude
5

Import dependencies

import 'dotenv/config';
import readline from 'node:readline';
import { Composio } from '@composio/core';
import { query, type Options } from "@anthropic-ai/claude-agent-sdk";

dotenv.config();
What's happening:
  • We're importing all necessary libraries including the Claude Agent SDK and Composio
  • The dotenv.config() function loads environment variables from your .env file
  • This setup prepares the foundation for connecting Claude with Deepgram functionality
6

Create a Composio instance and Tool Router session

async function chat() {
  const { COMPOSIO_API_KEY, USER_ID } = process.env;
  if (!COMPOSIO_API_KEY || !USER_ID) {
    throw new Error('COMPOSIO_API_KEY and USER_ID required in .env');
  }

  const composio = new Composio({ apiKey: COMPOSIO_API_KEY });

  // Create Tool Router session for Deepgram
  const session = await composio.create(USER_ID, {
    toolkits: ['deepgram'],
  });
  const mcpUrl = session?.mcp.url;
What's happening:
  • The function checks for the required COMPOSIO_API_KEY environment variable
  • We're creating a Composio instance using our API key
  • The create method creates a Tool Router session for Deepgram
  • The returned url is the MCP server URL that your agent will use
7

Configure Claude Agent with MCP

const options: Options = {
  permissionMode: 'bypassPermissions',
  mcpServers: {
    composio: {
      type: 'http',
      url: mcpUrl,
      headers: { 'x-api-key': COMPOSIO_API_KEY }
    }
  },
  systemPrompt: 'You are a helpful assistant with access to Deepgram tools via Composio.',
  maxTurns: 10,
};
What's happening:
  • We're configuring the Claude Agent options with the MCP server URL
  • permissionMode: 'bypassPermissions' allows the agent to execute operations without asking for permission each time
  • The system prompt instructs the agent that it has access to Deepgram
  • maxTurns: 10 limits the conversation length to prevent excessive API usage
8

Create client and start chat loop

const rl = readline.createInterface({
    input: process.stdin,
    output: process.stdout,
    prompt: 'You: '
  });

  console.log('\nChat started. Type "exit" to quit.\n');

  let isProcessing = false;

  async function ask(prompt: string) {
    isProcessing = true;
    rl.pause();

    process.stdout.write('Claude is thinking...');
    const stream = query({ prompt, options });

    let firstChunk = true;
    for await (const msg of stream) {
      const content = (msg as any).message?.content || (msg as any).content;
      if (Array.isArray(content)) {
        for (const block of content) {
          if (block.type === 'text' && block.text) {
            if (firstChunk) {
              process.stdout.write('\r\x1b[K');
              process.stdout.write('Claude: ');
              firstChunk = false;
            }
            process.stdout.write(block.text);
          }
        }
      }
    }
    process.stdout.write('\n\n');

    isProcessing = false;
    rl.resume();
    rl.prompt();
  }

  rl.on('line', async (line) => {
    if (isProcessing) return;

    const input = line.trim();
    if (input === 'exit') {
      rl.close();
      process.exit(0);
    }
    if (input) await ask(input);
    else rl.prompt();
  });

  await ask('What can you help me with?');
}
What's happening:
  • The readline interface is created to handle user input and output
  • The query function is used to send the user's input to the agent
  • The chat loop continues until the user types 'exit' or 'quit'
9

Run the application

try {
  await chat();
} catch (error) {
  console.error(error);
  process.exit(1);
}
What's happening:
  • The chat function is the entry point for the application
  • The try-catch block is used to handle any errors that occur

Complete Code

Here's the complete code to get you started with Deepgram and Claude Agent SDK:

import 'dotenv/config';
import readline from 'node:readline';
import { Composio } from '@composio/core';
import { query, type Options } from "@anthropic-ai/claude-agent-sdk";

async function chat() {
  const { COMPOSIO_API_KEY, USER_ID } = process.env;
  if (!COMPOSIO_API_KEY || !USER_ID) {
    throw new Error('COMPOSIO_API_KEY and USER_ID required in .env');
  }

  const composio = new Composio({ apiKey: COMPOSIO_API_KEY });
  const session = await composio.create(USER_ID, {
    toolkits: ['deepgram']
  });
  const mcp_url = session?.mcp.url;

  const options: Options = {
    permissionMode: 'bypassPermissions',
    mcpServers: {
      composio: {
        type: 'http',
        url: mcp_url,
        headers: { 'x-api-key': COMPOSIO_API_KEY }
      }
    },
    systemPrompt: 'You are a helpful assistant with access to Deepgram tools via Composio.',
    maxTurns: 10,
  };

  const rl = readline.createInterface({
    input: process.stdin,
    output: process.stdout,
    prompt: 'You: '
  });

  console.log('\nChat started. Type "exit" to quit.\n');

  let isProcessing = false;

  async function ask(prompt: string) {
    isProcessing = true;
    rl.pause();

    process.stdout.write('Claude is thinking...');
    const stream = query({ prompt, options });

    let firstChunk = true;
    for await (const msg of stream) {
      const content = (msg as any).message?.content || (msg as any).content;
      if (Array.isArray(content)) {
        for (const block of content) {
          if (block.type === 'text' && block.text) {
            if (firstChunk) {
              process.stdout.write('\r\x1b[K');
              process.stdout.write('Claude: ');
              firstChunk = false;
            }
            process.stdout.write(block.text);
          }
        }
      }
    }
    process.stdout.write('\n\n');

    isProcessing = false;
    rl.resume();
    rl.prompt();
  }

  rl.on('line', async (line) => {
    if (isProcessing) return;

    const input = line.trim();
    if (input === 'exit') {
      rl.close();
      process.exit(0);
    }
    if (input) await ask(input);
    else rl.prompt();
  });

  await ask('What can you help me with?');
}

try {
  await chat();
} catch (error) {
  console.error(error);
  process.exit(1);
}

Conclusion

You've successfully built a Claude Agent SDK agent that can interact with Deepgram through Composio's Tool Router.

Key features:

  • Native MCP support through Claude's agent framework
  • Streaming responses for real-time interaction
  • Permission bypass for smooth automated workflows
You can extend this by adding more toolkits, implementing custom business logic, or building a web interface around the agent.
TOOLS

Supported Tools

Every Deepgram action and event your agent gets out of the box.

Get Model by ID

Retrieve metadata for a specific Deepgram model by its UUID.

Get Public Models

Retrieve metadata on all public Deepgram models (speech-to-text and text-to-speech).

Get Project Balance

Retrieve one balance for a Deepgram project in its native units.

Get Project Billing Breakdown

Get a project's billing breakdown for an optional date range, preserving provider-reported amounts and units.

Get Project Request

Retrieve one request log for a Deepgram project by ID.

List Deepgram Projects

List Deepgram projects accessible to the API key.

Get Project Usage Breakdown

Get project usage grouped by provider usage fields, preserving native units.

Get Project Usage Summary

Retrieve aggregated project usage statistics, including speech durations, requests, channels, confidence and relevance, with usage breakdowns by model, accessor and tag.

List Project Balances

List all balances for a Deepgram project without converting their native amounts or units.

List Project Billing Fields

Return the fields available for project billing breakdowns.

List Project Purchases

List purchases for a Deepgram project without converting native values.

List Project Requests

List request logs for a Deepgram project with date and page controls.

List Project Scopes

Return a static list of Deepgram permission scope types and their descriptions.

List Project Usage Fields

Return the fields available for project usage breakdowns.

List Think Models

Tool to list available think models for AI agent processing and voice agent configuration.

Transcribe Pre-recorded Audio

Transcribe pre-recorded audio from a publicly accessible URL, returning the primary transcript in the first channel and alternative.

Text-to-Speech (REST)

Convert text to speech and return base64-encoded audio with its content type.

FRAMEWORKS

How to build Deepgram MCP Agent with another framework

Claude Code logo

Claude Code

Use Deepgram MCP with Claude Code

Claude Cowork logo

Claude Cowork

Use Deepgram MCP with Claude Cowork

OpenAI Agents SDK logo

OpenAI Agents SDK

Use Deepgram MCP with OpenAI Agents SDK

Codex logo

Codex

Use Deepgram MCP with Codex

OpenClaw logo

OpenClaw

Use Deepgram MCP with OpenClaw

Hermes logo

Hermes

Use Deepgram MCP with Hermes

Atomic Agent logo

Atomic Agent

Use Deepgram MCP with Atomic Agent

Google ADK logo

Google ADK

Use Deepgram MCP with Google ADK

LangChain logo

LangChain

Use Deepgram MCP with LangChain

Vercel AI SDK logo

Vercel AI SDK

Use Deepgram MCP with Vercel AI SDK

Mastra AI logo

Mastra AI

Use Deepgram MCP with Mastra AI

LlamaIndex logo

LlamaIndex

Use Deepgram MCP with LlamaIndex

CrewAI logo

CrewAI

Use Deepgram MCP with CrewAI

Pydantic AI logo

Pydantic AI

Use Deepgram MCP with Pydantic AI

AutoGen logo

AutoGen

Use Deepgram MCP with AutoGen

FAQ

Frequently asked questions

Start with Deepgram.It takes 30 seconds.

Managed auth, hosted MCP servers, and every Deepgram tool your agent needs.Free to start.

Start building