Usage Examples

These walkthroughs show what to say to an MCP client and which tool the client calls in response. The client picks tools and fills inputs from your prompt — you do not write the JSON yourself. The JSON below is what a reasonable call looks like, so you can see how a prompt maps to inputs. Tool names, full input lists, and response shapes are in Available Tools.

Screenshots

Generate a Chinese podcast:

Generate an English podcast:

Podcast from a topic

"Make a short English podcast about how LLMs changed search. Use two hosts."

The client first discovers speakers, then calls create_podcast with two speakerIds. create_podcast polls to completion and returns the finished episode, so you get the audio URL in one step.

{
  "tool": "create_podcast",
  "input": {
    "query": "How LLMs changed search",
    "speakerIds": ["sp_aria_en", "sp_leo_en"],
    "language": "en",
    "mode": "quick"
  }
}

For a head-to-head, two-host argument, ask for a debate. debate mode requires exactly 2 speaker IDs:

{
  "tool": "create_podcast",
  "input": {
    "query": "Is remote work better than office work?",
    "speakerIds": ["sp_aria_en", "sp_leo_en"],
    "language": "en",
    "mode": "debate"
  }
}

Podcast from a URL

"Turn this article into a podcast: https://example.com/post"

URLs and text passages go in sources. You can combine sources with a query to steer the framing.

{
  "tool": "create_podcast",
  "input": {
    "sources": ["https://example.com/post"],
    "query": "Focus on the practical takeaways",
    "speakerIds": ["sp_aria_en"],
    "language": "en",
    "mode": "deep"
  }
}

A single speaker ID produces a solo episode. Provide at least one of query or sources.

FlowSpeech narration

"Read this announcement out loud, word for word."

FlowSpeech is single-speaker narration. Use direct mode to read the source verbatim, or smart mode to let the model polish it first. create_flowspeech returns an episodeId and runs asynchronously — the client then polls get_flowspeech_status for the audio.

{
  "tool": "create_flowspeech",
  "input": {
    "sourceType": "text",
    "sourceContent": "This quarter we shipped three major features...",
    "speakerId": "sp_aria_en",
    "language": "en",
    "mode": "direct"
  }
}

To narrate a web page instead, set sourceType to url and put the link in sourceContent.

Script-first: review before audio

"Write the script for a podcast on prompt engineering, let me read it, then make the audio."

This is a two-step flow. First generate the script only:

{
  "tool": "create_podcast_text_only",
  "input": {
    "query": "An introduction to prompt engineering",
    "speakerIds": ["sp_aria_en", "sp_leo_en"],
    "language": "en",
    "mode": "quick"
  }
}

The client gets back an episodeId. Read the script with get_podcast_status, edit any lines, then generate audio. Pass your edits as customScripts so the audio matches your version:

{
  "tool": "generate_podcast_audio",
  "input": {
    "episodeId": "664e0c2b9f1a2b3c4d5e6f70",
    "customScripts": [
      { "speakerId": "sp_aria_en", "content": "Welcome back. Today: prompt engineering." },
      { "speakerId": "sp_leo_en", "content": "Let's start with what a prompt actually is." }
    ]
  }
}

Omit customScripts to generate audio from the script as-is.

Discover speakers

"What English voices are available?"

get_speakers returns the voice library. Filter by language to narrow it down. Each item carries a speakerId to feed into the generation tools.

{
  "tool": "get_speakers",
  "input": {
    "language": "en"
  }
}

Check your subscription

"Do I have enough credits to make a deep-dive podcast?"

get_user_subscription takes no inputs and returns your plan and credit balances. Call it before a long generation to confirm totalAvailableCredits covers it.

{
  "tool": "get_user_subscription",
  "input": {}
}

Next