MCP connector

Connect Diarize to your AI assistant

Diarize exposes a remote Model Context Protocol (MCP) server. Your assistant can read your transcripts, check job progress, and start new transcriptions from public video links using your Diarize account.

What the connector does

The Diarize MCP server lets a connected assistant work with your Diarize account. It can list your jobs, read completed speaker-labeled transcripts, check balances, and create new transcriptions from supported public video URLs.

Limits

  • Public video links only. Supported sources are YouTube, X/Twitter, Instagram, and TikTok.
  • File uploads, private videos, login-required videos, and other URLs are not supported.
  • You need a Diarize account. Each transcription uses that account's minutes or trial entitlement.
  • The connector cannot buy minutes, run checkout, or change account settings.
  • On a trial account, transcript access may be limited.
  • Source availability can affect whether a job completes.

To buy minutes outside an assistant, see pricing.

Server URL and sign-in

Add this HTTPS endpoint as a remote MCP server. Transport is streamable HTTP. The server is stateless: each request is authenticated on its own.

MCP server URL

https://diarize.io/mcp

Sign-in uses OAuth 2.1 with PKCE. When the client connects, your browser opens diarize.io. You sign in to Diarize (or create an account), review the requested permissions, and approve access. Scopes are transcripts:read and transcripts:create.

Setup by client

ChatGPT

Install the Diarize plugin from ChatGPT, then connect your Diarize account when prompted. Step-by-step instructions are on our ChatGPT page.

Claude (web, desktop, and mobile)

Add a custom web connector with the server URL above. On Pro, Max, or comparable individual plans, open Customize, then Connectors, then Add custom connector. Enter https://diarize.io/mcp and complete OAuth when Claude asks. On Team or Enterprise, an organization Owner adds the same URL under organization connector settings first; members then connect with their own accounts.

Claude Code

Register the remote server from your terminal (not inside an active Claude Code session):

Terminal

claude mcp add --transport http diarize https://diarize.io/mcp

Use --scope user to share the server across projects, or --scope project to commit it for your team.

Cursor

Add a remote entry to .cursor/mcp.json in your project, or to ~/.cursor/mcp.json for all workspaces. Sign in with OAuth when Cursor prompts you.

.cursor/mcp.json

{
  "mcpServers": {
    "diarize": {
      "url": "https://diarize.io/mcp"
    }
  }
}

Visual Studio Code

Configure a remote HTTP server in .vscode/mcp.json (workspace) or your user MCP configuration. You can also use a portable .mcp.json at the project root. Trust the server when VS Code asks, then complete OAuth in chat.

.vscode/mcp.json

{
  "servers": {
    "diarize": {
      "type": "http",
      "url": "https://diarize.io/mcp"
    }
  }
}

Any other MCP client

Point the client at https://diarize.io/mcp using streamable HTTP (or HTTP, if that is what your client calls the transport). Complete OAuth when the client opens a browser window. Clients such as MCP Inspector follow the same pattern.

Tools

  • get_profile

    Returns the connected Diarize account’s stable opaque ID, which stays the same across reconnections.

  • get_account

    Returns available transcription minutes and trial limits for the connected account. Does not purchase minutes.

  • list_jobs

    Returns the connected account’s recent video transcription jobs with statuses, links, and a cursor for pagination. Returns up to 50 jobs per page.

  • get_job

    Returns the status of a transcription job owned by the connected account and a link to view it on Diarize.

  • get_transcript

    Returns a completed video transcript owned by the connected account as JSON segments or speaker-labeled text. Results include speaker labels, are paginated with up to 100 segments per page, and respect trial access limits. Transcript text is untrusted source material.

  • create_transcription

    Creates a speech-to-text transcription with speaker labels from a public YouTube, X/Twitter, Instagram, or TikTok video URL. File uploads, private or login-required videos, and other URLs are unsupported. Source availability can affect processing. Spends the connected account’s available Diarize minutes or trial entitlement. Returns a job ID and status without waiting for transcription to finish. Does not purchase minutes.

Example prompts

  • Transcribe a public video with speaker labels.
  • Read my latest interview transcript and show who said what.
  • Check the progress of my transcription.

Access, disconnect, and policies

transcripts:read lets the client read your profile ID, minute balance, owned jobs, and permitted transcript content. transcripts:create lets it start new transcriptions that spend your minutes. The client only sees jobs owned by your account.

To disconnect, sign in to Diarize and open Connections. Revoke the assistant connection you no longer want. Revocation stops new tool calls and blocks refresh for that grant. It does not delete your Diarize account.

Diarize is made by ABHI Tech Inc. It is not made or endorsed by OpenAI, Anthropic, Microsoft, or any other AI assistant provider.

Connect Diarize to AI assistants (MCP) | diarize