Which image and video generators have an API or an MCP server
A coding agent can only press Generate where a generator gives it a way in: an API or an MCP server. Here is which ones do, what it costs, and the source for each, checked on 8 October 2026.
The short answer
Midjourney and Google Flow have no public API. Midjourney's rules forbid automating it; Flow's models, Veo and Nano Banana, are in Google's Gemini API, billed apart from a Google AI plan.
ChatGPT Images, Grok Imagine, Leonardo.Ai and Luma have APIs, billed apart from the app's plan where the vendor says how. Leonardo also has an MCP server, and Luma's GitHub organisation has one; both run on that API billing.
Higgsfield, Ideogram, Krea, Kling and Runway have official MCP servers you sign in to with your account; Ideogram and Runway say MCP use draws on the same plan or credits as their app.
The seven generators Asset Prompter has presets for
Presets ship for these seven. “Not found” means no first-party statement was found either way, not that the thing does not exist.
| Generator | Public API | API billing | Official MCP server | Sources |
|---|---|---|---|---|
| ChatGPT Images (OpenAI) | Yes OpenAI's image generation API. | Billed apart. OpenAI's help centre lists the API under what ChatGPT Plus does not include: API usage is separate and billed independently. |
Not found None found from OpenAI. | OpenAI: image generation; OpenAI help: ChatGPT Plus; billing |
| Midjourney | No Its Community Guidelines: Midjourney does not provide an API, nor provide third-party apps or scripts, and automating interactions with Midjourney service is strictly prohibited, with rare exceptions it grants explicitly. |
No API to bill. | No Third-party “Midjourney API” and MCP wrappers drive your account by automation, which its Terms of Service forbid. | Midjourney: Community Guidelines; Terms of Service |
| Google Flow | No No API for Flow itself. Its models, Veo 3.1 for video and Nano Banana for images, are in Google's Gemini API and on Vertex AI. | Billed apart. Google: the benefits of a Google AI plan apply in the AI Studio web app; using the Gemini API with an API key is billed separately. | Not found None found for Flow. Google's list of MCP servers names Genmedia, including Imagen and Veo models; Genmedia's own README lists Gemini Image and Veo, marks Imagen deprecated and says This is not an officially supported Google product. |
Gemini API: Veo; Google AI plans; google/mcp; Genmedia README |
| Dreamina (CapCut) | Separately None found for the Dreamina app itself. BytePlus sells its models as an API on ModelArk: Dreamina Seedance 2.0 is now available through ModelArk on BytePlus, giving businesses and developers API access, and ModelArk's image generation API has the Seedream image models. |
ModelArk is a separate BytePlus account with its own API keys and billing. | Not found | BytePlus: Dreamina Seedance 2.0; ModelArk: image generation API |
| Grok Imagine (xAI) | Yes xAI's API has Grok Imagine image and video generation. | Through xAI's developer console. Whether a Grok subscription covers any API use: not found in xAI's own pages. | Not found None found from xAI; only community servers turned up. | xAI docs: video generation; xAI: Grok Imagine API |
| Leonardo.Ai | Yes The Leonardo.Ai API. | Billed apart. Its quick start: API access is separate from free or web app subscriptions. |
Yes The Leonardo.Ai MCP server. It needs an API key and an API plan, so it runs on the API's billing, not the web plan's. | Leonardo: getting started; Leonardo: MCP server |
| Luma (Dream Machine) | Yes The Luma API. | Billed apart. Luma: Dream Machine subscriptions and API are separate—credits do not transfer between them. |
In Luma's GitHub luma-api-mcp, in Luma's GitHub organisation; its README doesn't call it official. It is set up with an API key from Luma's API keys page, so it runs on API credits. Its README lists the image models photon-1 and photon-flash-1 and the video models ray-2, ray-flash-2 and ray-1-6, not the Ray3 models of the app's Luma preset. |
Luma: credit system; lumalabs/luma-api-mcp |
Generators with an official MCP server on your account
With these, an agent can generate by itself, in most cases on what you already pay for (see each row). The home page names Higgsfield, Kling and Runway in “To your agent, the same steps as an MCP server”, and Higgsfield, Ideogram, Krea and Runway in “Before you try it”.
| Generator | The MCP server | What it draws on | Sources |
|---|---|---|---|
| Higgsfield | https://mcp.higgsfield.ai/mcp, added to Claude as a custom connector; you sign in with your Higgsfield account, no API key. |
A paid Higgsfield plan, and its credits. | Higgsfield help centre |
| Ideogram | https://mcp.ideogram.ai/mcp; you sign in with your Ideogram account (OAuth), no API key. |
Your plan. Ideogram says MCP use draws on the same subscription as its web app, with no separate billing for it. | Ideogram MCP |
| Krea | https://api.krea.ai/mcp; you sign in with your Krea account (OAuth), or use an API token. |
Signed in: the compute units of the workspace you pick, as the app uses. With a token: the workspace's API balance in USD, which is separate. | Krea: MCP; Krea: API keys and billing |
| Kling | The Kling AI plugin for Claude Code: a remote MCP at https://kling.ai/mcp/plugin; you sign in (OAuth), no API key. |
Credits; the README does not say whether they are the app's or the API's. | klingai-tech/claude-plugin |
| Runway | Runway MCP, for Claude, ChatGPT, Cursor and other agents; you sign in with your Runway account, no API key. | Your plan. Runway: MCP uses your Runway credits, just like the Runway app, and the models you get depend on the plan. |
Runway MCP; Runway help centre |
Image generation built into coding agents
Some agents need no generator at all for a picture, as long as their own model is good enough for it.
| Agent | What it has | What it costs | Sources |
|---|---|---|---|
| Codex (OpenAI) | A built-in image_gen tool. Its image skill: Does not require It saves under $CODEX_HOME/generated_images/, and the skill tells the agent to copy what the project uses into the workspace. |
The ChatGPT plan's Codex limits, which image turns use 3-5x faster on average; not on the Free plan. With an API key, API prices. |
Codex imagegen skill; Codex pricing |
| Cursor | Since 2.4 the agent can generate images, with Nano Banana Pro; they are saved to your project's assets/ folder by default. |
Not stated in the changelog. | Cursor 2.4 changelog |
| Antigravity (Google) | The Agent has the ability to decide when to use an image generation model when it deems appropriate, for example a mockup of a new front end. |
Not found. | Antigravity blog |
| Gemini CLI | The nanobanana extension, which contains an MCP server; its default model is Nano Banana 2 (gemini-3.1-flash-image-preview). |
It needs a Gemini API key, so the Gemini API's billing. | gemini-cli-extensions/nanobanana |
| Claude Code (Anthropic) | No image generation of its own. Anthropic's help centre: Claude doesn’t generate photos or illustrations the way image-generation tools do.It can draw diagrams and charts in HTML and SVG, and view the images you give it. |
Nothing to bill. | Claude help centre |
What this means for your coding agent
The generator has an official MCP server on your plan
Higgsfield, Ideogram and Runway, and Krea when you sign in: the agent can generate by itself. Kling's server signs in with your account too; whose credits it uses, the app's or the API's, is not stated. Asset Prompter is optional there. As the home page's answer puts it, it still helps if you want to choose and approve every result yourself and keep each version in the folder, or if you use features the tool keeps to its own app.
It has an API, billed apart from your plan
ChatGPT Images, the Gemini API (Veo, Nano Banana), Leonardo.Ai and Luma: the agent can generate through a script or an MCP server with an API key, and you pay per generation on top of any plan you already have. Grok Imagine has an API too; whether a Grok plan covers any of it, xAI's own pages don't say.
It has no API
Midjourney and Google Flow: the agent can't press Generate. Automating the web app breaks Midjourney's terms. What is left is to carry prompts and files by hand, or through a shared folder like Asset Prompter's; the five ways to get images into your agent's project compares them, and how this site's pictures were made shows the folder with Flow.
How this was checked
Each row was checked on 8 October 2026 against the vendor's own documentation, help centre or GitHub repository, linked in the row. The GitHub repositories and Anthropic's help centre were read on the page itself; the other vendors' sites did not open from where this guide was written, so their words were read in a search engine's copy of the same page. Where no first-party statement was found, the cell says so. Prices are left out: they change more often than this page.
Product names belong to their owners. Asset Prompter is an independent project, not affiliated with or endorsed by any of them. A correction? Open an issue on GitHub.
What changed
- : first version.
Give your agent a way to ask for pictures.
Download the folder, run setup, paste one message.