# Connect Interrupt

You are setting up Interrupt for your user. Follow these steps in order.

Interrupt makes private podcasts from what the user wants to learn (links, PDFs, notes, a YouTube video or a podcast episode), in a voice they pick, and keeps them in their own library to listen to. It is a remote MCP server (Streamable HTTP) that signs the user in with OAuth 2.1. There is no API key.

- MCP server URL: https://turr.fm/mcp
- Server name: `interrupt`

## If you are a coding agent that can add MCP servers

Claude Code, Codex, Cursor, VS Code, Gemini CLI and similar. Add the server with the line for your client:

- Claude Code: `claude mcp add --transport http interrupt https://turr.fm/mcp`
- Codex: `codex mcp add interrupt --url https://turr.fm/mcp`
- Gemini CLI: `gemini mcp add --transport http interrupt https://turr.fm/mcp`
- Cursor: add this to `~/.cursor/mcp.json` (or open cursor://anysphere.cursor-deeplink/mcp/install?name=interrupt&config=eyJ1cmwiOiJodHRwczovL3R1cnIuZm0vbWNwIn0=):

```json
{
  "mcpServers": {
    "interrupt": {
      "url": "https://turr.fm/mcp"
    }
  }
}
```

- VS Code: add this to `.vscode/mcp.json` (or open vscode:mcp/install?%7B%22name%22%3A%22interrupt%22%2C%22type%22%3A%22http%22%2C%22url%22%3A%22https%3A%2F%2Fturr.fm%2Fmcp%22%7D):

```json
{
  "servers": {
    "interrupt": {
      "type": "http",
      "url": "https://turr.fm/mcp"
    }
  }
}
```

Then:

1. Start the sign-in (your client may need a restart or `/mcp` to pick up the server). If you handle OAuth yourself: call the server without a token, follow `resource_metadata` in the 401 response to the authorization server and register there (dynamic client registration). The user signs in to Interrupt and chooses **Allow**; give them the sign-in link if it doesn’t open by itself.
2. Verify: call `list_episodes` and show the user their library (it may be empty).
3. If your client can’t add MCP servers or open an OAuth sign-in, don’t retry: tell the user, and send them to https://turr.fm/connect.

## If you are a chat assistant (Claude.ai or ChatGPT)

You can’t add tools to yourself from a chat. Walk the user through this in their app’s settings, then ask them to come back and say “done”.

Claude (web, desktop or phone):

1. Open **Customize → Connectors**, tap **+**, then **Add custom connector**.
2. Paste the URL below and tap **Add**.
3. Tap **Connect**, sign in and choose **Allow**.

ChatGPT:

1. In **Settings → Apps → Advanced**, turn on **Developer mode**.
2. Tap **Create**, name it Interrupt, paste the URL below and choose **OAuth**.
3. Sign in and choose **Allow**. Custom connectors need a paid ChatGPT plan.

The URL to paste: https://turr.fm/mcp

When they say “done”, the Interrupt tools are available in a new chat (and sometimes this one). Try `list_episodes`. If you can’t see the tools here, ask them to start a new chat and ask you there.

## Once connected

Two ways to make audio. Both return at once with a `web_url` to share; the work runs in the background, and `get_episode` says when it’s ready.

- **make_podcast**: a podcast about the material, written for one host. Sources are links (web pages, PDFs, a YouTube video or a Spotify podcast episode: the podcast is then about it) and/or text.
  - style: `summary` (“Catch me up”), `explainer` (“Teach me”), `deep_dive` (“Go deep”), `story` (“Tell it as a story”) or `critique` (“Challenge it”)
  - length_minutes: 5, 10, 15, 20 or 30
  - optional: instructions, voice (the name of one of the user’s own voices; leave out for the standard voice), tone (`natural`, `warm`, `calm`, `excited` or `serious`), language (`auto`, the default, follows the sources; or `en` English, `zh` Chinese, `ja` Japanese, `es` Spanish, `ar` Arabic, `fr` French, `de` German, `pt` Portuguese or `ko` Korean)
- **Read it aloud**, word for word: `add_link` with a web page or PDF link, or `add_text` with text you have (a page behind a login, notes). Optionally a voice and a tone.
- Also: `list_episodes`, `get_episode`, `update_episode` (title, description, artwork) and `delete_episode` (confirm with the user first).

What to tell the user they can now do:

- **Make a podcast.** From articles, PDFs and notes, or about a YouTube video or a podcast episode: in the style, length, voice and tone you ask for.
- **Read it aloud.** A web page, a PDF or pasted text, word for word, in the standard voice or one of yours.
- **Edit details.** Titles, descriptions and artwork.
- **Delete items.** Free up space. Deleted items lose their Q&A too.
- **See your library.** What you have, what you’ve finished, where you left off.

Things they can ask you:

- “Make me a 10-minute explainer podcast from these three articles, in my voice.”
- “Make a 5-minute podcast about this talk: youtube.com/watch?v=…”
- “Read this article aloud in my voice, calm: …”
- “Delete everything I’ve finished, then make me a podcast from this article for my drive.”