How to generate images and video from Claude, Cursor and ChatGPT with one MCP server
Claude, ChatGPT and Cursor are great at text and code. On their own, Claude and Cursor can't draw a picture or render a video clip. The Model Context Protocol (MCP) fixes that: you plug a remote server into your agent, a
Claude, ChatGPT and Cursor are great at text and code. On their own, Claude and Cursor can't draw a picture or render a video clip. The Model Context Protocol (MCP) fixes that: you plug a remote server into your agent, and the agent gets new tools it can call in the middle of a conversation.
In this post I connect one remote MCP server, Seedara, to the agents I use and show what the workflow looks like. Disclosure: I work on Seedara. Connecting is free, and the free plan covers images and audio; video needs a paid plan.
What the agent gets
After setup the agent has nine tools:
| Tool | What it does |
|---|---|
generate |
Starts a generation; images come back right in the chat |
get_generation |
Status and result of a long run (video, music) |
list_models |
Image, video and audio models with prices |
get_model |
A model's settings: formats, duration, quality |
get_price |
The exact price before the run |
get_balance |
How many credits are left |
list_generations |
Your latest work with links |
upload_file |
Your own photo, video or recording, by link or as a file |
list_elements |
Your saved characters, locations, products and voices |
Behind them are 140+ models from 19+ developers: Kling, Veo and Seedance for video, Nano Banana, GPT Image and FLUX for images, ElevenLabs and Suno for voice and music. You can name a model yourself or just describe the result and let the agent pick one.
Step 1. Copy the server URL
https://seedara.ai/api/mcp
It is a Streamable HTTP server with OAuth. There is no API key: the first time the agent connects, a browser window opens, you sign in and press Allow.
Step 2. Add it to your agent
Claude (claude.ai or the desktop app)
- Open Customize β Connectors.
- Press + β Add custom connector, name it
Seedara, paste the URL and press Add. - Press Connect next to Seedara, sign in and press Allow.
- In a chat press + β Connectors and turn Seedara on.
The free Claude plan allows one custom connector.
Claude Code
claude mcp add --transport http seedara https://seedara.ai/api/mcp
Start Claude Code, type /mcp, pick seedara and sign in in the browser.
Cursor
Put this into ~/.cursor/mcp.json (or .cursor/mcp.json inside a project):
{
"mcpServers": {
"seedara": { "url": "https://seedara.ai/api/mcp" }
}
}
Turn seedara on and sign in when Cursor opens the window. If it doesn't connect, Output (Ctrl+Shift+U) β MCP Logs shows the reason.
VS Code
Run MCP: Add Server from the Command Palette (Ctrl+Shift+P), or put the config into .mcp.json at the project root. VS Code asks you to sign in on first start; the tools appear in Copilot Chat under Configure Tools.
ChatGPT
This works on chatgpt.com on paid plans.
- Settings β Security and login, turn on Developer mode (some accounts already have it).
- Open Plugins, press + β Create app β Create MCP App.
- Name
Seedara, the server URL, authentication OAuth. Create it, sign in and press Allow. - In a new chat press + and pick Seedara.
In some clients, Claude and ChatGPT especially, the connection shows up after 5β10 minutes. If the tools don't appear, wait and start a new chat.
Step 3. Ask for something
Prompts that work well:
- "Make three cover options for a morning-coffee video, 16:9. Tell me the price first."
- "Remove the background from this photo and make it white, for no more than 20 credits: https://example.com/photo.jpg"
- "Which models make video with sound? Compare their price for 5 seconds."
- "Animate the latest picture in my library: slow push-in, 5 seconds."
- "Write a 30-second jingle for a coffee shop, warm and acoustic."
Here is what a call looks like in Claude:
You: Draw a blonde by a pink sports car at a night gas station, like a film still.
Claude: seedara Β· generate(model: "nano-banana-2-new",
prompt: "blonde by a pink sports car at a night gas station, film still",
params: {"size": "16:9"})
Done β nano-banana-2-new (image).
Result: https://seedara.ai/api/files/β¦
Images come back in the chat. Video and music take from a minute to a few: the agent gets a generation id and checks get_generation later by itself, or you ask it to.
Why do this from an agent
-
The price first. With
get_pricethe agent tells you the cost before it runs, and you can set a cap ("no more than 20 credits"). - Assets while you code. In Claude Code or Cursor you can ask for an icon, a cover or a hero image for the project and get a file link right next to your code.
- Compare models. Ask for the same prompt in two or three models and keep the best one.
-
Your own references.
upload_filetakes your photo or recording by link, andlist_elementskeeps the same character or product across shots.
What the agent can and can't do
The agent signs in with your account through OAuth. It can make generations on your credits and see models, prices, your balance and your generations. It can't pay or top up, can't see your password, e-mail or payment methods, and can't change account settings. A failed generation refunds the credits. You can disconnect any agent in one click on the MCP page; changing your password disconnects all of them.
Cost
Connecting is free. A generation through an agent costs the same as on the website. The free plan gives 100 credits every 30 days for images, audio and text; video starts with paid plans.
Instructions for Windsurf / Devin, Codex, Gemini CLI and other clients: https://seedara.ai/mcp
If you try it, tell me in the comments which agent you connected and what didn't work. I read them.
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes β full credit and traffic to the original publisher.