# AI Product Shots with Claude MCP: A Real Kive Session (2026) > Connect Kive's MCP connector to Claude: the documented setup, the route from product photos to checked product shots, and one recorded session's costs. Published: 2026-08-25 Updated: 2026-10-14 Author: Olga Stogova Tags: claude-mcp, kive-mcp, model-context-protocol, ai-product-shots, ai-automation Canonical: https://kive.ai/learn/claude-mcp-product-shots Use the Kive connector in Claude when you want an assistant to turn a saved product into product shots and short videos without opening the Kive app. You connect from Kive's MCP settings, create the product from its photos, remix an idea or generate with a studio, then compare every result against the source photo before you keep it. The evidence is Kive's MCP documentation plus one session Kive ran for this article, which used a character model rather than a product. For a whole catalog in one pass, use Bulk generate in the Kive app instead: the connector has no batch tool, and Kive has no public REST API for generation. ## Key takeaways - The connector covers the product route: create a product from photos or a product page, remix an idea or use a studio, then check it. - Setup starts in Kive under Settings → MCP → Claude → Connect, and needs available credits, Claude custom-connector access and permission to authorize. - Kive's docs put each idea remix image at 10 credits; in the recorded session, each still made with a saved character model also cost 10. - A video needs a saved product or model, because a brief alone cannot start one, and the connector's video tool takes no studio. - There is no batch tool and no public REST API for generation, so catalog-scale runs belong in Kive's Bulk generate. --- ## What the Kive connector gives Claude The Kive MCP server exposes 31 tools ([kive.ai/mcp/tools](https://kive.ai/mcp/tools)), and Claude reads that list when it connects. The ones that matter for product shots map onto Kive's own objects: saved products built from your photos, character models that carry one person across images, studios that save a scene's lighting and framing, video presets for motion, and style references that carry a look into new frames. According to [Kive's MCP overview](https://kive.ai/docs/mcp/overview), a connected client can browse all of these, create a saved product from a product page, image URL or upload, remix an idea with your product, generate images and videos, and edit existing images. The practical difference from Kive's composer is that nothing needs clicking. In the app you reference a saved model by typing @modelname and pick a studio from a menu. Here you write a sentence, such as shoot my ceramic mug in a clean studio at 4:5, and Claude looks the names up with the find tools before it calls a generation tool. The [ChatGPT MCP guide](/learn/chatgpt-mcp-product-visuals) walks the same tool loop from the other assistant's side. What Claude draws on its own, without any connector, is covered in [can Claude generate images](/learn/can-claude-generate-images). How a chat assistant driving tools differs from a creative agent that plans the set and asks before it spends is covered in [what a creative AI agent for ecommerce does](/learn/creative-ai-agent-ecommerce). ### Where the credits come from Anything you make through Claude uses the same credits and plan as the Kive app ([Claude MCP page](/mcp/claude)). `generate_product_image` and `generate_product_video` can return a credit estimate instead of spending anything, which is worth asking for before a run of several images. `edit_image` has no estimate option, so edits are the calls to watch on a tight balance. Model training also uses credits ([credits FAQ](https://kive.ai/docs/troubleshooting-faqs/credits-and-limits-faq)), and saved product and character models are available on paid plans ([train a custom model](https://kive.ai/docs/creating-with-ai/train-a-custom-model)). ### MCP is not a REST API Kive doesn't publish a REST API for generation. The MCP server at `https://mcp.kive.ai/mcp` is the programmatic route, and an MCP client drives it: Claude, ChatGPT, or tools such as Zapier and n8n that accept a custom connector ([connect any client](https://kive.ai/docs/mcp/connect-any-client)). There is no batch tool either. To shoot a catalog in one scene, Claude starts one `generate_product_image` call per product with the same studio and polls each job. Kive's [Bulk generate](https://kive.ai/docs/creating-with-ai/bulk-generate) runs a set of products through one Studio in a single pass, on paid plans and on desktop. --- ## Connect Kive to Claude Setup starts in Kive, not in Claude. The [connect-to-Claude docs](https://kive.ai/docs/mcp/connect-to-claude) list four prerequisites: a Kive account with access to the workspace you want to use, available Kive credits, access to Claude custom connectors, and permission to authorize Claude to use your Kive account. ### Connect Kive to Claude **Open MCP settings in Kive** Open your avatar menu and go to Settings → MCP. **Click Connect on the Claude row** Under Apps, find the Claude row and click Connect. Kive opens Claude's custom connector setup with the Kive details filled in. The [Claude MCP page](/mcp/claude) has the same connect button. **Authorize Kive** Follow the prompts. Kive asks you to allow access before Claude can use your workspace, and it still enforces workspace access, plan permissions and credit availability through MCP. **Confirm with a read-only request** In Claude, ask: Show my Kive workspaces. Claude should list the workspaces your account can reach. Nothing generates, so nothing is spent. If you add the connector by hand in Claude's settings instead, the server URL is `https://mcp.kive.ai/mcp`, which you can also copy from Settings → MCP → Custom connections. To disconnect, click Revoke on the Claude row under Settings → MCP → Apps; Kive's docs call that the authoritative disconnect ([manage connections](https://kive.ai/docs/mcp/manage-and-troubleshoot)). --- ## The product route: from photos to checked shots This is the documented route for a product, taken from Kive's [MCP overview](https://kive.ai/docs/mcp/overview) and [tool list](https://kive.ai/mcp/tools). It is not the session recorded further down, which used a character model, so read it as documented capability rather than a measured run. ### Product shots in Claude with Kive **Create the product from photos** Ask Claude to add the product to Kive from its product page link, an image URL or uploaded photos. In the Kive app, a product model takes one to four photos of the same product from different angles. Use sharp photos: Kive's docs warn that when text or logos aren't sharp in the inputs, the model tends to invent garbled versions ([train a custom model](https://kive.ai/docs/creating-with-ai/train-a-custom-model)). **Pick an idea or a studio** Ask for ideas for the product. `find_ideas` returns ideas matched to it, with the price of a remix and your credit balance, and `remix_idea` recreates the one you pick with your product. A remix runs as a Kive Agent chat that may ask a question or two first, and each remix image uses 10 credits. To shoot it in a studio instead, ask Claude which studios suit the product, then generate with `generate_product_image` and ask for an estimate first. **Check the product against the source photo** Open each result at full size next to the source photo and compare the label text, logo, shape and color. Reject any image where one of them changed and run it again; for a single wrong detail, ask Claude to fix it with `edit_image`. If the same detail keeps drifting, Kive's docs suggest editing the product to use one sharp photo from its best angle and adding an instruction such as "maintain exact logo". **Make a clip from the saved product** `generate_product_video` needs a saved product or model, because a brief alone cannot start a video. One request accepts up to four video presets and up to four sampled outputs, and video starts with one clip with audio on by default. Kive's docs list logos, buttons and colors as the details that can drift, with a fix list of their own, so the check in step three covers every image, not a sample. --- ## The anchor workflow, run for real What follows is one session Kive ran in Claude for this article. It used a saved character model, not a product, for a fictional Scandinavian-minimal look, so it shows how the tool loop and its costs behaved in one run; it is not a product-fidelity test. The problem it tackles is the one that breaks prompt-only tools: scene two never quite matches scene one. The fix is to make one image the anchor and derive everything else from it. ### Step 1: one hero still The session started with a single request to generate the hero look on a saved character model: a full-body editorial still in a bright minimal studio, wearing an oversized sand trench over a cream crewneck and ivory trousers. One model-anchored generation, 10 credits, ready in about 44 seconds. That image became the anchor for the rest of the session; it is also this article's hero image, unretouched. ### Step 2: the same look, every scene Each scene call then reused three things: the same character model, a studio resolved by name, and the hero still passed back as a style reference, with the garments described identically in every brief. Two scenes ran this way, one at blue hour on a wet street and one against a warm ochre backdrop. The trench, knit and trousers came back matching in color and construction in both, on the same woman, 10 credits each. Repeating the garment wording matters as much as the reference image: the style reference carries the look, the words pin the details. `generate_product_image` accepts a saved product in place of a character, but this session did not test one. Training and reusing the character itself is covered in the [character-model docs](https://kive.ai/docs/guides/ai-character-models), and our [character consistency guide](/learn/same-model-across-ai-product-photos) covers where the approach still breaks. _Scene two: same outfit, resolved into the Blue Hour scene – Kive studio: Blue Hour Portrait_ --- ## Adding motion with video presets Video runs through its own tool, and its inputs differ from the image tool in two ways. `generate_product_video` needs a saved product or model, since a brief alone cannot start a video, and it has no studio input, even though studios can style video inside the Kive app ([generate a video](https://kive.ai/docs/creating-with-ai/generate-a-video)). It takes a motion preset resolved by name, such as Pushing in, Holding product or Revealing slowly, plus a brief; to carry a still scene's atmosphere into the clip, describe it in that brief. Kive's app renders clips of 4 to 15 seconds, and video cost varies with the Draft, Premium or Max mode ([plans and credits](https://kive.ai/docs/account-workspace-settings/plans-and-credits)). In this session the estimate for a 5-second, 9:16 Premium clip with audio came back at 100 credits; no clip was rendered. Against the session's 10-credit stills, that is the cost of ten frames, so lock the look on stills first and spend video calls only on looks that already passed review. _Scene three: the anchor look against the Ochre backdrop – Kive studio: Ochre Studio Portrait_ --- ## What the session cost Every number below came back from the tool calls in that one session. The video row is an estimate, not a render. | Call | Credits | Time to ready | |---|---|---| | Hero still (character model) | 10 | ~44 s | | Scene still (model + studio + style reference) | 10 each | ~70 s each | | 5 s Premium video with audio | 100, estimate only | not rendered | | Image or video generation with estimateOnly | 0 | instant | The three stills totaled 30 credits. Saved character and product models are on paid plans, so this workflow starts on Basic at $40 a month for 2,000 credits ([kive.ai/pricing](https://kive.ai/pricing)). One caution transfers from every asynchronous pipeline: generation returns a job, not an image, and Claude polls until the render reports ready, so a job stalled for several minutes is worth reissuing rather than waiting on. --- Run the product route once on your own product before you scale it: one product, one idea or studio, and every result checked against the source photo. You make the creative calls and the accept-or-reject decision. Claude does the bookkeeping, resolving names, writing briefs, watching jobs and keeping the description identical from frame to frame. Start from the [Claude MCP page](/mcp/claude) or the [connect-to-Claude docs](https://kive.ai/docs/mcp/connect-to-claude). For catalog refreshes and seasonal variants, the [ChatGPT MCP guide](/learn/chatgpt-mcp-product-visuals) maps the larger shapes, and [bulk image generation for a product catalog](/learn/bulk-image-generation-product-catalogue) covers the in-app route for a whole catalog. ## FAQ ### How do I connect Kive to Claude? Start in Kive: open Settings → MCP, find the Claude row under Apps and click Connect. Kive opens Claude's custom connector setup with its details filled in. Authorize access, then ask Claude to show your Kive workspaces to confirm it works. You need available Kive credits, access to Claude custom connectors and permission to authorize. ### What does an image cost through the Kive MCP? It comes out of the same workspace credits and plan as the Kive app. Kive's docs put each idea remix image at 10 credits, and in the session this article records, each still made with a saved character model also cost 10. The image and video tools can return an estimate without spending anything, so ask Claude for one first. ### Can Claude keep the same model and outfit across scenes? In this article's one recorded session, yes. Generate one hero still with a saved character model, then pass that image back as a style reference into every scene call while describing the garments identically in each brief. The same coat, knit and trousers came back in three scenes. Check every frame anyway: studios give a consistent style, not identical output. ### Does the Kive MCP generate video too? Yes. generate_product_video needs a saved product or character model, because a brief alone cannot start a video, and it takes no studio, so describe the scene in the brief. One request can use up to four video presets. Cost varies by quality mode; in this article's session, a 5-second Premium clip with audio was estimated at 100 credits. ### Does Kive have an API for product shots? Not a public REST API for generation. The MCP server at https://mcp.kive.ai/mcp is the programmatic route, and an MCP client drives it: Claude, ChatGPT, or automation tools such as Zapier and n8n. It has no batch tool, so Claude runs one generation per product. For a whole catalog in one pass, use Bulk generate in the Kive app.