Overview

The Carbon Voice MCP Server implements the Model Context Protocol (MCP) to expose Carbon Voice functionality to AI assistants and other MCP-compatible clients. It bridges the Carbon Voice API (voice messages, conversations, folders, workspaces, and AI actions) into the MCP model so tools like Cursor and Claude Desktop can access voice-centric actions and context as part of a model-assisted workflow.

You can run the server in two modes: a remote HTTP-hosted MCP that uses OAuth2 for authentication (zero local prerequisites), or a local stdio MCP that runs via npx and authenticates with a Carbon Voice API key. The server follows MCP security guidelines (OAuth 2.1, HTTPS, secure sessions, input validation, and rate limiting) so it is suitable for both local development and production integrations.

Features

Message management: create, list, and fetch voice messages, conversation messages, and direct messages
Conversation and participant management: read and manipulate conversations and members
Folder operations: create, move, and list folders and their contents
Workspace administration: fetch workspace metadata and settings
User operations: search and retrieve user details
AI actions: run prompts and retrieve AI-generated responses within Carbon Voice
Attachment support: add link attachments to messages
Security best-practices: OAuth 2.1 support, HTTPS-only endpoints, session and input safeguards

Installation / Configuration

Two transport options: HTTP (remote) and stdio (local). HTTP is recommended for simple setup and better security.

Remote HTTP (recommended)

No local binaries or API keys required
Uses OAuth2 — users authenticate through Carbon Voice

To use the remote server from MCP clients, point the MCP server URL to:

https://mcp.carbonvoice.app

Local stdio (developer / offline)

Requires Node.js (npx) and a Carbon Voice API key
Useful for local development, debugging, or non-OAuth workflows

Example: run via npx with an environment variable for the API key:

export CARBON_VOICE_API_KEY="your_api_key_here"
export LOG_LEVEL="info"    # optional: debug, info
npx -y @carbonvoice/cv-mcp-server

Cursor configuration examples

Remote (HTTP) — add to Cursor settings:

{
  "mcpServers": {
    "Carbon Voice": {
      "url": "https://mcp.carbonvoice.app"
    }
  }
}

Local (stdio) — add to Cursor settings:

{
  "mcpServers": {
    "Carbon Voice": {
      "command": "npx",
      "env": {
        "CARBON_VOICE_API_KEY": "your_api_key_here"
      },
      "args": ["-y", "@carbonvoice/cv-mcp-server"]
    }
  }
}

Claude Desktop configuration

Remote (HTTP) — add a custom connector with Remote MCP Server URL:

https://mcp.carbonvoice.app

Local (stdio) — add similar npx-based server entry in the Claude Desktop config file:

{
  "mcpServers": {
    "Carbon-Voice": {
      "command": "npx",
      "env": {
        "CARBON_VOICE_API_KEY": "your_api_key_here"
      },
      "args": ["-y", "@carbonvoice/cv-mcp-server"]
    }
  }
}

Environment variables (stdio only)

CARBON_VOICE_API_KEY — required for local stdio
LOG_LEVEL — controls logging: debug | info (default: info)

Available Resources

Source code / releases: https://github.com/PhononX/cv-mcp-server
Carbon Voice API docs: https://api.carbonvoice.app/docs
Model Context Protocol spec & security guidance: https://modelcontextprotocol.io
Support / security contact: [email protected]

Use Cases

Build a voice-enabled assistant that composes and sends voice messages on behalf of users. The assistant can list recent messages, synthesize replies (via Carbon Voice AI actions), and attach links or metadata.
Integrate conversation context into a chatbot workflow: fetch conversation history, summarize threads for model prompts, and post AI-generated replies into the correct conversation or folder.
Workspace administration tools: create and organize folders programmatically, move message items between folders, and audit participants and workspace settings.
Local development: run the stdio server with an API key to iterate quickly without provisioning OAuth clients, then switch to the remote HTTP transport for production rollouts.
Hybrid flows for automation: trigger AI prompts (transcription, summarization) on new voice messages and surface results back into Carbon Voice as messages or attachments.

Notes for Developers

Prefer the HTTP transport for production deployments — it leverages OAuth2 and requires no local secrets.
Use the stdio transport when you need a fast local loop or your environment cannot perform OAuth.
The server adheres to MCP security best practices (TLS-only endpoints, secure session tokens, input validation, and rate limiting). Follow the same principles when integrating into your client to maintain a secure end-to-end flow.

Carbon Voice MCP Server for AI Voice Assistants