Claude Turns Product Links Into Editable AI Ad Workflows
ElevenLabs presents its MCP for Claude as a way to manage AI ad production through chat while retaining access to the underlying ElevenCreative Flows workflow. In a tutorial demonstrated by Alec Wilcock, users can start with a product link, approve stills and voice takes, revise motion or narration, and reformat a finished ad, while Flows exposes the separate image, video, voice and composition steps. The process still depends on user decisions—and on manual product details when a linked page cannot be retrieved.
Chat is the approval layer; Flows is the production layer
Claude serves as the place to give direction, review output, and request changes in ordinary language. ElevenCreative Flows is the underlying production canvas, accessible through an Open flow link when the workflow itself needs inspection or adjustment.
A user can begin with a product URL and a request for a cinematic ad, then make creative decisions in chat: narrator, mood, product details, still-image approval, motion changes, and delivery format. The process shown is staged rather than a single opaque generation. It plans the voiceover and scenes, produces stills for approval, then creates video and compiles the result.
Before it goes and generates the entire thing, it builds out a structure of what the voiceover should look like, what each scene should look like.
The Flows canvas makes that production chain visible. In the crew-sweatshirt project, GPT Image 2 receives a product-photography instruction for a dark charcoal sweatshirt floating against a near-black studio background, lit from behind for a premium fashion-editorial look. Gemini Omni Flash receives the motion direction: a slow cinematic dolly-in as the garment rotates in zero gravity. Separate visible nodes cover narration, voice synthesis through Eleven Multilingual v2, and composition.
The canvas therefore exposes more than a completed file. It separates image prompting, video prompting, narration, voice generation, and composition into editable parts, while Claude remains the interface for routine creative direction.
Creative direction can be revised without rebuilding the ad
The ElevenLabs MCP connects Claude to image, video, voice, sound, and music capabilities. Alec Wilcock identifies GPT Image 2 and Nano Banana 2 for images, plus Seedance 2.5 and Gemini Omni Flash for video. The on-screen model list also includes Krea 2, Recraft v4, MiniMax H3, and FLUX.1.
The material point is not merely model access, but the kinds of decisions a user can alter through follow-up instructions.
| Control | What happens in Claude or Flows |
|---|---|
| Narrator | Claude returns playable voice takes and lets the user search for and select another voice. |
| Voice model and delivery | A follow-up instruction requests more emotion using Eleven v3 and audio tags while retaining the selected voice. |
| Creative direction | The product-ad workflow asks the user to choose a mood for the planned scenes. |
| Product detail | Claude asks the user to supply the garment colorway when it cannot load the store page. |
| Image approval | Four generated stills are presented for review before the work moves into video. |
| Motion revision | After the ad is assembled, the user asks to make the product rotation quicker. |
| Placement format | The completed ad is reformatted into a vertical TikTok version from the existing project context. |
The voiceover exercise provides a compact version of this workflow. Wilcock asks Claude to create an ad voiceover for a sparkling-water brand called Core. Claude fetches voices and returns several playable takes using Adalina — Professional & Friendly and eleven_multilingual_v2. The audible sample repeats: “Some things don't need to be complicated.”
Wilcock then selects Change voice, searches the library for something “more moody and deep,” and chooses Aida — Sultry, Deep and Silky. The MCP regenerates the takes with Aida. He next instructs Claude to give the voiceover more emotion using Eleven v3 and audio tags; Wilcock says the request rewrites the prompt, changes the model, and keeps the selected voice.
That sequence treats voice choice, performance, and model choice as revisions to an existing asset rather than settings that must be settled before generation begins.
A product link supplies context, but it does not eliminate missing inputs
The fullest example begins with the ElevenLabs store’s 11_11 Crew Sweatshirt, listed on screen at $74.99. Wilcock gives Claude a product URL and asks for a cinematic ad. He says that providing the link tells the system to fetch assets from the website, after which it builds the ad’s voiceover and scene structure.
The URL does not guarantee that every needed detail arrives. Claude says it could not load the store page and asks the user to identify the sweatshirt’s colorway, offering black, cream/off-white, and grey as choices. Product context still has to be supplied when retrieval fails.
The workflow also asks the user to choose a narrator and mood. Its available voice choices include George, a British storyteller; Adrian, a deep American voice; Sarah, a confident female voice; and Charlie, an energetic Australian voice. For tone, it offers moody and premium, warm and cozy, or futuristic tech.
After those inputs, Claude produces four still images for approval. The interface then reports a finished 15-second, 1080p cinematic ad: a black 11_11 Crew Sweatshirt in moody rim light with Adrian’s voiceover. The completed video carries lines including “Built different,” “Worn with attitude,” “Made for those who create,” and “Create, wear, repeat.”
The project remains editable after assembly. A request to “Make the rotation quicker” follows the completed ad, and the final horizontal piece can be reframed through a further prompt: “Can you also make this vertical please for TikTok?”
Connection requires Claude’s connector settings and an ElevenLabs account
To authorize the integration, Wilcock goes through Claude’s Settings menu to Connectors, selects Add, then Browse connectors, and searches for ElevenLabs. The ElevenLabs MCP can then be added through the plus sign; after onboarding, the user clicks Connect and signs in to an ElevenLabs account.
Once connected, the source positions Claude alongside existing workspace connections for documents, websites, and email, while adding access to the ElevenCreative Flows agent and the available media-generation tools. The operational premise is that creative work can begin from the same chat where a user already has relevant context, then move into the Flows canvas when the underlying production chain needs direct attention.