Product Ad
Copy product prompt
You can always select and copy the text below. Press Ctrl+C or ⌘C to copy manually.
Connect Kinovi to the assistant you already use. Ask for an image, a clip, or audio, and see the live models and the credit quote first.
In ChatGPT on the web, open Plugins, choose +, then Add custom MCP server.
Open ChatGPT guideName it Kinovi. Set authentication to OAuth. No API key.
https://kinovi.ai/api/mcp
You can always select and copy the text below. Press Ctrl+C or ⌘C to copy manually.
Create the plugin, approve Kinovi in the browser, then type @ and pick Kinovi.
Use Claude Code in the project where you want Kinovi available.
Open Claude Code guideRun this command.
claude mcp add --transport http kinovi https://kinovi.ai/api/mcp
You can always select and copy the text below. Press Ctrl+C or ⌘C to copy manually.
Run /mcp, choose kinovi, then Authenticate. Approve the browser page.
Go to Customize → Connectors, then Add → Custom → Web.
Open Claude Desktop guideName it Kinovi. Choose Register automatically. Do not use Claude's published identity.
https://kinovi.ai/api/mcp
You can always select and copy the text below. Press Ctrl+C or ⌘C to copy manually.
Choose Connect and approve Kinovi in the browser. Then enable it under + → Connectors.
In Cursor, open Settings → Tools & MCP, or Customize, and open mcp.json.
Open Cursor guideMerge this into mcp.json and save. Leave your other servers in place.
{
"mcpServers": {
"kinovi": {
"url": "https://kinovi.ai/api/mcp"
}
}
}You can always select and copy the text below. Press Ctrl+C or ⌘C to copy manually.
Authenticate kinovi in the browser. Then ask Cursor to list models and show a quote.
Use the Codex CLI in the project where you want Kinovi available.
Open Codex guideRun this command.
codex mcp add kinovi --url https://kinovi.ai/api/mcp
You can always select and copy the text below. Press Ctrl+C or ⌘C to copy manually.
Run codex mcp login kinovi and approve the browser page. Then ask for a model and a quote.
Add a remote Streamable HTTP server in the client you already use.
Open the MCP pageLeave the client ID blank if the app asks. Kinovi registers it.
https://kinovi.ai/api/mcp
You can always select and copy the text below. Press Ctrl+C or ⌘C to copy manually.
Approve the browser page, then ask for list_models before any generation.
Connect Kinovi MCP to this MCP client using browser OAuth. Read https://kinovi.ai/docs/mcp first. Check that this client supports remote Streamable HTTP, OAuth 2.1 with PKCE S256, and dynamic client registration (DCR). If not, stop and recommend a supported client; do not fall back to an API key or a local proxy. If an authorized local configuration editor is available, configure only the kinovi remote URL using this client's documented schema. Otherwise give me the manual UI steps. Do not copy another client's JSON schema blindly. After I complete browser authorization, enable the server's tools or reload the chat as required. Preserve all other MCP servers and settings. Reuse an existing kinovi entry; do not create duplicates. If that entry has a different URL or custom authentication, explain the conflict and ask before changing it. Never display existing credentials. The endpoint is https://kinovi.ai/api/mcp, using remote Streamable HTTP and browser OAuth. Do not configure an API key, Authorization header, command, or args for this remote connection. I must complete browser sign-in and approve Kinovi consent myself. Never approve consent for me or ask me to paste passwords, tokens, or API keys into chat. After setup, call only Kinovi's read-only list_models tool to verify the connection. A configuration entry, connected indicator, or OAuth label alone is not proof. Confirm success only after list_models returns successfully. Do not call create_task, upload files, generate media, or spend credits during setup. If editing, authorization, or tool access is unavailable or denied, stop, explain the remaining manual steps, and do not claim success or bypass restrictions. For future generation, first read get_model_params, estimate_cost with the exact proposed inputs, and get_account. Show the estimated credits and wait for my explicit approval before create_task. Suggest a whole-number maxCredits cap for my approval; it caps the submission charge, not later usage adjustments. Save the returned taskId and use get_task to resume an interrupted run before submitting another task.
You can always select and copy the text below. Press Ctrl+C or ⌘C to copy manually.
Connected? Start by listing models.
“Use Kinovi to list image and video models with their prices. Do not create a task yet.”
Listing models is free. Confirm the quote before you generate. Menus depend on the app and plan.
You can always select and copy the text below. Press Ctrl+C or ⌘C to copy manually.
Your assistant lists live models, checks the quote, and resumes the same task if the chat drops.
Ask which image, video, or audio models are available, and what each one accepts.
list_modelsget_model_paramsGet the credit price for the exact inputs, and your balance, before anything is submitted.
estimate_costget_accountApprove once. If the chat drops, check the same task instead of starting another.
create_taskget_taskGenerate images, videos, and audio with live catalog models. Call each capability on its own or combine them in your workflows.
You can always select and copy the text below. Press Ctrl+C or ⌘C to copy manually.
You can always select and copy the text below. Press Ctrl+C or ⌘C to copy manually.
You can always select and copy the text below. Press Ctrl+C or ⌘C to copy manually.
You can always select and copy the text below. Press Ctrl+C or ⌘C to copy manually.
Pick a live video model, read the duration and resolution, and quote the credits before anything runs. This ten-second garage spot is four hard cuts, made with Seedance 2.5 Flash.
You can always select and copy the text below. Press Ctrl+C or ⌘C to copy manually.




The same image model can explore a place, a time of day, and a material without leaving the chat. These four stills were made with Midjourney V8.
You can always select and copy the text below. Press Ctrl+C or ⌘C to copy manually.
Ask for a video, image, or audio model. Kinovi lists what is live, then quotes it before anything runs.
Your assistant runs the steps. Kinovi supplies the models, the quote, and the result.
Bring a reference into the chat you already use. Kinovi quotes the clip before it runs.
list_models → get_model_paramsestimate_cost → your approvalcreate_task → get_task“Make a short cinematic clip from this image. Show me the quote before anything runs.”
Use Kinovi to find a live video model that accepts my reference image. Read its parameters and estimate a short cinematic clip. Show the quote and wait for my approval. Submit only once after approval, save the taskId, and check it with get_task if the conversation is interrupted.
You can always select and copy the text below. Press Ctrl+C or ⌘C to copy manually.
Use Kinovi to find an available image model for a clean studio photo of a ceramic mug. Read its parameters, suggest suitable settings, and show the estimated credits. Wait for my approval before creating any task.
You can always select and copy the text below. Press Ctrl+C or ⌘C to copy manually.
Use Kinovi to find available audio generation models. Explain their supported inputs, compare suitable options, and estimate the cost for my brief. If no model fits, say so. Do not create a task until I approve the model, settings, and estimated credits.
You can always select and copy the text below. Press Ctrl+C or ⌘C to copy manually.
maxCredits limits the first charge. It is not a guarantee on the final usage bill.Connection, credits, file uploads, and what happens after you submit.
It connects ChatGPT, Claude, Cursor, or Codex to your Kinovi account. Add the Kinovi URL, sign in through the browser, and ask for images, video, or audio. You do not run a local server.
No. Menus, file uploads, and previews depend on the app, the version, and your plan. ChatGPT custom connectors are not on every account. Follow the guide for the app you have.
No. Connecting, listing models, reading parameters, estimates, and task checks do not spend credits. create_task charges the normal API price when you submit. Website-only free entitlements do not apply.
Ask for the quote and wait. Approval is a conversation rule, not a server-enforced confirmation dialog. A whole-number maxCredits value rejects a submission whose first charge is higher. It is not a guarantee on the final usage bill. There is no cancel tool yet.
Yes, when that model accepts them. Pass a public URL, or have the app upload the file. Requesting an upload URL alone does not upload the file. The app must send the bytes and get a success response before it uses the URL. Uploads last 7 days and the public URL stops working after deletion. Uploading needs write access.
Keep the task ID and check that task first. A dropped chat does not mean the job failed. Starting again can be a second charge.
No. Add the Kinovi URL and sign in through your browser. API keys are only for a separate server-to-server setup. Never paste a key or token into the chat.
Revoke a client from API Keys → Connected apps. Access stops there. Current setup steps and tool details stay in the MCP docs.