Berner Setterwall4 min read

    Make a Spotify Audio Ad From Your Agent, in One Call

    An audio ad is thirty seconds of someone talking, usually over music. Writing the script is the easy part, and your agent already does it well. Getting from the script to a file an ad platform accepts has meant a voice actor or a studio tool, and an export with the right loudness. Now your agent can hand you the file.

    create_audio_ad is live on Cogny's MCP endpoint. Give your agent the script, word for word, and it comes back with one MP3 in your Creative library:

    • A voice-over that says your words exactly, in the language the script is written in.
    • An instrumental music bed under it, if you describe one. Leave it out for voice only.
    • Mixed to Spotify's audio ad spec: up to 30 seconds, 44.1 kHz stereo, 192 kbps, -16 LUFS.

    TL;DR

    • Your words, exactly. Up to 700 characters. About 76 spoken words fit 30 seconds with music, 79 without.
    • Pick the voice by name or by description: "warm female, British", "deep male, American". The result lists the other voices, so a second version can name one.
    • Describe the music in a line: mood, tempo, instruments. It is always instrumental and sits low under the voice.
    • Checked against the script. Cogny transcribes the take and compares it with your words. Every number and price must match.
    • Loudness measured, not assumed. The result reports the finished file's loudness.
    • The images Spotify asks for, too. Add companion_image and you also get a square companion image in your brand and your logo on a square tile, both at the size Spotify takes.
    • Cost: 1 credit voice only, 2 with music, 1 more for the companion image, charged when the ad is delivered. Every plan, paid from credits.
    • In Cogny's chat and by voice too: ask the chat or the voice agent in the app for an audio ad and it makes the same one.

    The voice says what you wrote

    A wrong price in an audio ad is worse than a typo: nobody can see it, and it plays to everyone. So every take is transcribed and compared with the script before it is mixed.

    Numbers, prices, units and web addresses are spoken in full ("$280" becomes "two hundred and eighty dollars", "cogny.com" becomes "cogny dot com"), and each number in the script has to come back in the transcript, in order. If a take doesn't match, Cogny records it once more and keeps the better one. If it still differs, you get the ad with a warning that quotes what the voice actually says, so you can decide.

    The word count works the same way. A script is measured as it is said, so "$280 a month" counts as seven words, not three. If the script can't fit the length, the call is refused before anything is recorded, with how many words to cut.

    What your agent sends

    {
      "script": "Still paying $280 a month for marketing connectors? Cogny connects Google Ads, Meta and sixty more to your own AI for $9 a month. Start free at cogny.com.",
      "voice": "warm female, British",
      "music": "upbeat, modern, light electronic with a warm bass"
    }
    

    That script is 39 words as spoken and comes back as an ad of about 14 seconds. The ad is as long as the script takes to say; it is never padded to 30 seconds. seconds sets a shorter limit if the placement needs one (10 to 30).

    A client that supports MCP Tasks gets a task to poll, and a confirmation prompt before credits are spent. Any other client gets a task_id for get_audio_ad, and cancel_audio_ad stops it. An ad that fails or is cancelled before delivery costs nothing.

    Getting it onto Spotify

    Spotify asks for two pictures with every audio ad: a companion image shown while it plays, and your logo. Each is square, at least 600×600, JPEG or PNG, 200 KB at most. Add "companion_image": true to the call (or describe the picture you want) and the result carries both, ready to use: a 640×640 image generated in your brand, and your own logo from Brand & Context centred on a plain tile. The logo is never redrawn by a model.

    Cogny does not put the ad on Spotify by itself. Download the three files from the result or the Creative library and add them to an ad in Spotify Ads Manager.

    What it costs

    Credits
    Voice only1
    With music2
    Companion image and logo1 more
    A long script recorded twice, with music3 at most

    It is charged once, when the ad is delivered, and needs 3 credits available to start. Free workspaces pay from the 50-credit welcome gift, Solo from top-up credits, and Cogny AI, Cogny Pro and Cloud from the monthly credits.

    Try it

    Connect your agent to app.cogny.com/mcp and ask it:

    "Write a 20-second audio ad for our autumn sale, read it back to me, then make it with a calm British voice and soft piano."

    The tool reference is in the MCP docs.