Image and video generation in Claude over MCP

The short version: Claude does not draw pictures or make video on its own — it works with text and can look at images, but not create them. What Claude can do is call tools. Connect a generator as an MCP server and, when you ask it to "draw", Claude will write the prompt itself, call the generation and hand back a link to the finished file. The same works in Claude Code, Cursor, LM Studio and any other MCP client.

How it works

You write to Claude in plain language: "make a cover for an article about remote work, 16:9, calm tones".
Claude turns the request into a text-to-image tool call — with a detailed prompt and format parameters.
The server generates the image, stores the file and returns its description and a link to Claude.
Claude shows you the link and can carry on: "make it brighter" is now an image-to-image call on the file from the previous step.

The strength of this pairing is precisely step 2. A good prompt for a generator is a skill of its own, and Claude writes one better than most people: it accounts for style, composition, light and for what you said five messages ago.

Try the engine before setting anything up

Before connecting the server to Claude, you can see how the same image generator behaves right here, in the Twelver mini-chat:

Loading…

Which engines you can connect

Each engine is its own MCP server with its own tools. The URLs look like https://twelver.app/api/mcp/v1/image-generation; the full list is in the builder below.

Server	What it does	When to take it
Nano Banana	Image from text, editing an image, merging up to 5 images, describing an image	The all-round starting point: fast, inexpensive, understands edits phrased in words
Seedream	From text and from a reference, blending several images, image series	Series in one consistent style, photorealism
Ideogram	Generation with legible text, transparent PNG, remix, background removal	Lettering on the image, logos, stickers
Seedance	Video from text, from an image, multimodally; editing and extending	The main choice for clips
Hailuo	Video from text, an image, a pair of frames and references, with sound	A clip between two given frames
HappyHorse	Video from text, an image, references; editing a video by instruction	Editing a finished clip in words
Veo, Wan, Grok	Video from text and image; Veo and Wan with sound	When a specific engine is required

What every plan includes, and what needs PRO

Nano Banana, Seedance, Hailuo and HappyHorse are available on any plan, the entry one included. The Seedream, Ideogram, Grok (images and video), Veo, Wan sets are available on PRO and above.

How to connect it to Claude

In claude.ai and Claude Desktop, as a custom connector: Settings → Connectors → Add custom connector, the server URL, sign in. In Claude Code, by command:

claude mcp add --transport http twelver-image https://twelver.app/api/mcp/v1/image-generation
claude mcp add --transport http twelver-video https://twelver.app/api/mcp/v1/seedance-video-generation

Then /mcp and sign in through the browser. All the details, including signing in with a key, are in MCP in Claude. For other clients: Cursor, LM Studio, Codex, OpenCode.

No keys: paste the URL into your client and sign in — Claude, Cursor and any other OAuth client.

Open Claude → Settings → Connectors

Add custom connector → paste the URL

Build your server

Check the tools — URLs and config update themselves.

Connector URLs

Image Generation (Nano Banana)

https://twelver.ru/api/mcp/v1/image-generation

Video Generation (Seedance)

https://twelver.ru/api/mcp/v1/seedance-video-generation

What comes back, and how to work with it

A generation returns a link to a file, not a picture embedded in the answer. Claude puts the link in its message — click it and the image or clip opens in the browser. The link is permanent: you can forward it or paste it into a document.

Along with the link the server returns a file id. That is what the next steps in the same conversation use: "make the background darker" (image from image), "bring this image to life" (video from an image) — Claude passes the id to the right tool itself. Which is why refining is easier in one conversation than in a fresh one.

Files without a wordmark come after the first purchase

On the starting tokens images open with a small wordmark on them — quite enough to judge the result and tune the prompt. After the first purchase files are available in the original and can be downloaded.

Parameters worth naming to Claude

Aspect ratio — square, 16:9, 9:16, 4:3 and others. If you do not name one, you get a square.
Resolution — 1K by default. 2K and 4K are noticeably dearer; ask for them when the picture is going to print or onto a large screen.
For video, the duration and the camera movement. A short clip is cheaper and faster; start there and extend what worked.

Slow generations: video and timeouts

An image is ready in seconds, a video in minutes, and the tool call stays open the whole time. Some clients have a shorter call timeout: in Claude Code you raise it with the timeout field on the server in .mcp.json, in Codex with tool_timeout_sec. If the client reports a timeout, do not repeat the request straight away — the generation on the server may have finished.

What it costs

The servers draw on the shared Twelver token balance. For a sense of scale, here is what the Basic plan at $8 a month buys if you spend all of it on one thing:

≈ 851K images through Nano Banana

≈ 49480p videos through Seedance

Figures as of 6 September 2026; current prices are on the pricing page.

Besides generation, it is worth connecting web search to Claude — together they cover jobs like "find what the brand's logo looks like and make a banner in its style". What MCP is in general is in the overview of MCP servers.

Frequently asked questions

Can Claude generate images?

On its own, no: Claude works with text and can analyse images, but not create them. Generation appears once you connect a generator to Claude as an MCP server (in claude.ai, through a custom connector). Claude then writes the prompt itself, calls the generation and returns a link to the finished image.

How do I create video in Claude?

Connect a video-generation MCP server — Seedance or Hailuo, for instance — as a connector in claude.ai, or with claude mcp add --transport http in Claude Code. After that Claude makes clips from text, or from a picture generated earlier in the same conversation. Generation takes minutes, so in Claude Code it is worth raising the tool timeout.

Which MCP server should I choose for image generation?

To start with, Nano Banana: fast, inexpensive and good at understanding edits phrased in words. For lettering on the image and for logos, Ideogram; for photoreal shots and people who will later move in video, Seedream. Each engine is a separate server, so they can be combined.

Where do I see the picture Claude generated?

The answer carries a link to the file — follow it and the image opens in the browser. Alongside the link comes a file id, which Claude uses to keep editing the picture or turn it into video in the same conversation.

Can I edit my own picture over MCP?

The editing tools work with files that are already on the server — first of all with the results of earlier generations in the same conversation. The convenient pattern is to generate a base and refine it with requests like "remove the person on the left" or "make it evening instead of day".

How much does image generation over MCP cost?

It depends on the engine and the resolution. For a sense of scale: the Basic plan at $8 a month is about 85 1K images through Nano Banana, or about 49 480p clips through Seedance. 2K and 4K cost noticeably more.

Try it yourself

Everything in this guide runs inside Twelver

One chat for text, images, video, music and voice — no separate services or subscriptions.

Open Twelver chat

Related pageMCP Toolsets

Rate your experience