Carbon Voice MCP Server

What is Carbon Voice MCP Server?

The Carbon Voice MCP Server is the official remote MCP server for Carbon Voice, an async voice messaging app for teams and AI agents. It lets Claude, Cursor and other MCP clients read and send voice messages, manage conversations, folders and action items, and run AI actions such as summaries.

Carbon Voice is built around one-tap voice messages that are transcribed automatically, with AI summaries and action items on top. Its makers, PhononX, publish the MCP server as open source (PhononX/cv-mcp-server) and host the remote version at https://mcp.carbonvoice.app, which uses OAuth 2.1 so you never handle an API key.

How to use Carbon Voice MCP Server?

For the remote server you only need a Carbon Voice account. Per the README, the HTTP version "runs entirely in the cloud and uses OAuth2 authentication", so no API key or Node.js install is required. The first time you use it, your client opens a browser window where you sign in to Carbon Voice and grant permissions.

https://mcp.carbonvoice.app

Claude Code

claude mcp add --transport http carbon-voice https://mcp.carbonvoice.app

Run /mcp and choose Authenticate to complete the Carbon Voice sign-in.

Claude Desktop and claude.ai

From the README: open Search and Tools → Manage Connectors, click Add custom connector, name it "Carbon Voice", enter https://mcp.carbonvoice.app, save, then click Connect and sign in with your Carbon Voice account.

Cursor

The README's remote configuration:

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

Save and restart Cursor; it guides you through OAuth on first use.

VS Code

In .vscode/mcp.json:

{
  "servers": {
    "carbon-voice": {
      "type": "http",
      "url": "https://mcp.carbonvoice.app"
    }
  }
}

A local stdio version (npx -y @carbonvoice/cv-mcp-server with a CARBON_VOICE_API_KEY) also exists, but its API key must be requested from Carbon Voice's developer team, so the remote server is the easier option.

Key features of Carbon Voice MCP Server

• Official server from the Carbon Voice team, hosted remotely with OAuth 2.1.
• Message management: list, read and send conversation messages and direct messages.
• Voice memos created from text (spoken via text-to-speech) or from audio at a public HTTPS URL.
• Conversation, user and workspace lookups.
• Folder operations: create, rename, move and delete folders, and file messages into them.
• AI actions: list and run Carbon Voice AI prompts on messages, including conversation summaries.
• Action items: create, assign and track them, and extract suggested items from messages with AI.
• Inbox, unread and heard-state searches, share links and link attachments.

Use cases of Carbon Voice MCP Server

• Catching up: "Summarize the unheard messages in my team conversation."
• Hands-off updates: "Send a voice memo to the design channel saying the review moved to Thursday."
• Task capture: "Pull action items out of today's messages from Sarah and add them to my list."
• Organizing: "Move all messages about the Q3 launch into a new folder called Launch."
• Status check: "What action items are assigned to me and still open?"

Troubleshooting Carbon Voice MCP Server

• The client shows no tools. The remote server needs an OAuth sign-in before it lists tools; complete the browser flow or re-authenticate from the client's MCP settings.
• "Missing Authorization header" when testing the URL. This is expected for unauthenticated requests; use an MCP client that supports OAuth.
• A voice memo from a URL fails. The server only fetches public https audio URLs, refuses private or loopback addresses, and caps file size (25 MB by default).
• List results are very long. Most read tools accept an optional response_fields list to return only the fields you need.