The Kive MCP server exposes 28 tools for creating product images and video, importing products, searching a workspace, and organising the results. Each one is listed below by its exact tool name.
| Tool | What it does |
|---|---|
| generate_product_image | Generates product images from a saved product, a saved model, or a plain brief. Can return a credit estimate instead of spending anything. |
| generate_product_video | Generates product video from saved products and models, optionally driven by a video preset. |
| edit_image | Edits an existing Kive image from a natural-language instruction, keeping the rest of the frame intact. |
| generate_prompt_from_image | Reads one image and writes the prompt that would recreate its look, for reuse across other products. |
| create_product_in_kive | Creates a product from exactly one source: an e-commerce product page URL, an image URL, a local file, or a completed upload handle. |
| create_product_image_upload_url | Issues an upload URL for one local product image, for cases where the agent holds the file rather than a link. |
| confirm_product_image_upload | Confirms an upload finished so the image can be attached to a product. |
| show_product_image_upload_widget | Fallback that renders a browser upload widget when a direct upload is blocked. |
| create_character_in_kive | Creates a reusable character model from reference images of the same person, so the same face carries across shots. |
| list_workspaces | Lists the Kive workspaces the connected account can reach, with their ids. Most flows start with this call. |
| find_products | Searches saved products by name, category or description. |
| find_models | Searches saved AI models: characters and trained brand styles as well as products. |
| find_generations | Searches images and videos already generated in the workspace. |
| find_boards | Lists boards and the assets saved on a given board. |
| find_studios | Lists or searches studios: the published Kive catalog plus the workspace’s own custom studios. |
| recommend_studios | Returns studios matched to a product that is still being created, as a visual contact sheet to choose from. |
| find_video_presets | Lists video presets available for product-video generation. |
| check_image_generation_status | Checks one image generation without rendering it. For headless use. |
| check_video_generation_status | The same status check for a video generation. |
| check_character_creation_status | Polls a character-model creation job until it is ready. |
| show_kive_image | Renders a finished image inline in chat. This is the tool that owns final image cards. |
| show_kive_video | Renders a finished video inline in chat. |
| create_kive_board | Creates an empty board, for a moodboard or a campaign. |
| update_kive_board | Renames a board, edits its description, or moves it under another board. |
| move_kive_board_items | Moves saved images or videos from one board to another. |
| kive_ping | Checks that the MCP server itself is reachable. |
| kive_graphql_status | Checks that the server can reach the Kive API behind it. |
| check_kive_auth_status | Checks the identity the server is authenticated as. Called only after another tool reports an auth error. |
The four tools that produce something new. Each one starts a durable job, so an assistant can begin a render and collect it later.
Generates product images from a saved product, a saved model, or a plain brief. Can return a credit estimate instead of spending anything.
Inputs: Required: workspaceId, plus a subject — productId, productQuery, modelIds or brief. Optional studioId, aspectRatio, samples, mode, styleImageUrls and estimateOnly.
Ask for it like this: “Shoot my ceramic mug in the Cream studio, 4:5, one image.”
Generates product video from saved products and models, optionally driven by a video preset.
Inputs: Required: workspaceId, plus a saved model — productId, productQuery or modelIds. A brief alone cannot start a video. Optional videoPresetId, duration, aspectRatio, withAudio and estimateOnly.
Ask for it like this: “Turn that hero shot into a five-second product clip.”
Edits an existing Kive image from a natural-language instruction, keeping the rest of the frame intact.
Inputs: Required: workspaceId and instruction, plus the source image as generationId, outputId or imageUrl.
Ask for it like this: “Make the background warmer and remove the shadow.”
Reads one image and writes the prompt that would recreate its look, for reuse across other products.
Inputs: Required: imageUrl, a publicly reachable https image URL.
Ask for it like this: “Describe the style of this reference so I can reuse it.”
Kive generates from saved products rather than from prompts alone, so most workflows start here. A product can come from a store URL, an image URL, or a file upload.
Creates a product from exactly one source: an e-commerce product page URL, an image URL, a local file, or a completed upload handle.
Inputs: Required: workspaceId, plus exactly one source — productUrl, imageUrl(s), imageFilePath(s) or uploadedImageHandle(s). Optional productName and brief.
Ask for it like this: “Add this product to Kive: <your product page URL>”
Issues an upload URL for one local product image, for cases where the agent holds the file rather than a link.
Inputs: Required: workspaceId, fileName and contentType. Optional productName and brief.
Ask for it like this: “Upload these three photos of the bottle as one product.”
Confirms an upload finished so the image can be attached to a product.
Inputs: Required: workspaceId, uploadedImageHandle and the upload’s httpStatus.
Ask for it like this: “Used automatically after an upload completes.”
Fallback that renders a browser upload widget when a direct upload is blocked.
Inputs: Nothing required. Pass the fallbackUploadWidget reference returned by create_product_image_upload_url.
Ask for it like this: “Used automatically when a direct upload fails.”
Creates a reusable character model from reference images of the same person, so the same face carries across shots.
Inputs: Required: workspaceId, plus reference images of one person as imageUrl(s), imageFilePath(s) or uploadedImageHandle(s). Optional characterName.
Ask for it like this: “Make a character model from these photos of our founder.”
Read-only lookups. An assistant asked to show a library uses these and never touches a generation tool.
Lists the Kive workspaces the connected account can reach, with their ids. Most flows start with this call.
Inputs: Nothing required.
Ask for it like this: “Which Kive workspaces do I have?”
Searches saved products by name, category or description.
Inputs: Required: workspaceId. Optional query.
Ask for it like this: “Show me the fragrances in my catalog.”
Searches saved AI models: characters and trained brand styles as well as products.
Inputs: Required: workspaceId. Optional query and modelType.
Ask for it like this: “List my trained brand styles.”
Searches images and videos already generated in the workspace.
Inputs: Required: workspaceId. Optional assetType of IMAGE, VIDEO or ALL.
Ask for it like this: “Show the videos I generated last week.”
Lists boards and the assets saved on a given board.
Inputs: Required: workspaceId. Optional boardQuery or boardId.
Ask for it like this: “What is on my spring campaign board?”
Studios are saved scenes that bundle lighting, framing and environment. They are how output stays consistent without prompt engineering.
Lists or searches studios: the published Kive catalog plus the workspace’s own custom studios.
Inputs: Required: workspaceId. Optional creative-direction query.
Ask for it like this: “Find a clean e-commerce studio for skincare.”
Returns studios matched to a product that is still being created, as a visual contact sheet to choose from.
Inputs: Required: jobId from product creation, and workspaceId.
Ask for it like this: “Which studios suit this product?”
Lists video presets available for product-video generation.
Inputs: Nothing required. Optional query.
Ask for it like this: “What video presets can I use?”
Generation is asynchronous. These tools separate checking a job from rendering its output, so headless pipelines and chat surfaces can use the same server.
Checks one image generation without rendering it. For headless use.
Inputs: Required: generationId and workspaceId. Optional waitMs.
Ask for it like this: “Is that render finished yet?”
The same status check for a video generation.
Inputs: Required: generationId and workspaceId. Optional waitMs.
Ask for it like this: “Is the clip done?”
Polls a character-model creation job until it is ready.
Inputs: Required: the character creation jobId, and workspaceId.
Ask for it like this: “Is my character model trained?”
Renders a finished image inline in chat. This is the tool that owns final image cards.
Inputs: Nothing required. Pass generationId or jobId with workspaceId, or a direct imageUrl.
Ask for it like this: “Show me the result.”
Renders a finished video inline in chat.
Inputs: Nothing required. Pass generationId with workspaceId, or a direct videoUrl.
Ask for it like this: “Play the clip here.”
Boards are where generated work is kept and shared, so an agent can file its own output rather than leaving it loose.
Creates an empty board, for a moodboard or a campaign.
Inputs: Required: workspaceId and title. Optional parentBoardId to nest it.
Ask for it like this: “Make a board called Autumn Launch.”
Renames a board, edits its description, or moves it under another board.
Inputs: Required: workspaceId and boardId, plus the fields being changed — title, description, parentBoardId or moveToTopLevel.
Ask for it like this: “Rename that board to Autumn Launch 2026.”
Moves saved images or videos from one board to another.
Inputs: Required: workspaceId, boardItemIds and targetBoardId.
Ask for it like this: “Move these three shots to the launch board.”
Diagnostics, used when something is not responding. They are not part of a normal creative flow.
Checks that the MCP server itself is reachable.
Inputs: Nothing required. Optional message to echo back.
Ask for it like this: “Is the Kive MCP server up?”
Checks that the server can reach the Kive API behind it.
Inputs: Nothing required.
Ask for it like this: “Can the connector reach Kive?”
Checks the identity the server is authenticated as. Called only after another tool reports an auth error.
Inputs: Nothing required.
Ask for it like this: “Which Kive account is this connector using?”
Individual tools are the vocabulary. These are the sentences – the call sequences an agent runs to finish a real piece of work. Each step links to the tool it uses.
You say: “Shoot everything in my Kive catalogue in the same clean studio, 4:5.”
You get: One image per product, all in the same scene, rendered in the chat as they finish.
You say: “Give me four ad variants of the ceramic mug, each in a different scene.”
You get: Four takes on the same product in different studios, ready to test as creative.
You say: “Make a five-second clip of the mug using one of the video presets.”
You get: A rendered product clip, playable in the chat.
You say: “Add this product to Kive and show me what it looks like in a few scenes.”
You get: A saved product in the workspace, plus first shots in studios matched to it.