What it does
Visual Sandbox is a remote MCP server for making media. One endpoint gives an
agent more than a hundred models across five kinds of output: image, video,
audio, 3D and vector.
The agent reads the catalogue, reads a model's input fields, checks the price,
starts the run and collects the result. Finished images come back as image
content, so the model can judge what it made instead of guessing from a URL.
Connect
{
"mcpServers": {
"visual-sandbox": {
"type": "http",
"url": "https://visualsandbox.com/mcp"
}
}
}
Your client opens a browser once to authorize. There is no API key to paste.
Tools
| Tool | What it does |
|---|
list_models | The catalogue, with the price of each call |
get_model | Every input field of one model, and what it accepts |
estimate_cost | What a call would cost, before you make it |
generate | Start a run and charge the account |
get_job | Progress, then the output URLs and the cost |
list_jobs | Recent runs on the account |
upload_media | Put a local file on the CDN to use as a reference |
list_presets / get_preset | Saved model-and-settings recipes |
get_balance | Credit left, and the markup the prices carry |
top_up | A checkout link for adding credit |
list_skills / get_skill | Written guidance on prompting each family of model |
Every tool carries a read-only hint, so your client knows which ones spend
money before it calls them. Only generate does.
Pricing
You pay for each generation: the provider cost plus a small platform fee,
written on a receipt you can read. There is no subscription and no monthly
minimum. estimate_cost tells you the price before the run, and get_job
reports what the run actually cost.
Skill packs
Prompt craft changes the result more than the choice of model does. The server
carries the written guidance as skill packs, and get_skill reads one. An
agent that reads the image pack before its first prompt produces noticeably
better work.
Links
Comments