Claude Image Generation

Visual Sandbox runs image, video, audio, 3D and vector models behind one MCP server.

CommunityOnlineNo sign-inGlobalFreeRead-only

What it can do

List Models: List the models this account can run, with the price per call. Filter by modality to see only what makes images, video, audio, 3D or vector. Call this before generate when the model is no
Get Model: Everything about one model: what every input field means, which fields are required, the accepted values, and the price. Read this before the first generate with a model, because inputs dif
Estimate Cost: What one call to a model would cost this account, in US dollars, with the supplied inputs. Partial inputs give a partial answer rather than an error. Nothing is charged and no job is cr

What data it sees

Do you need an account

No: the server works without sign-in

Visual Sandbox runs image, video, audio, 3D and vector models behind one MCP server. Your agent lists the catalogue, reads a model's inputs, estimates the cost, starts a generation and collects the result. You pay per generation, with no subscription.

Finished images come back as image content, so the model can judge what it made instead of guessing from a URL. Skill packs carry the written guidance on prompting each family of model — read one with get_skill before the first prompt.

Only generate spends money, and it is annotated so your client knows before it calls.

Server tool list (13)

Raw names from tools/list. Only developers need these.

list_models	List the models this account can run, with the price per call. Filter by modality to see only what makes images, video, audio, 3D or vector. Call this before generate when the model is not already known — a slug invented from memory will not resolve.
get_model	Everything about one model: what every input field means, which fields are required, the accepted values, and the price. Read this before the first generate with a model, because inputs differ from model to model.
estimate_cost	What one call to a model would cost this account, in US dollars, with the supplied inputs. Partial inputs give a partial answer rather than an error. Nothing is charged and no job is created.
generate	Start a generation and charge this account for it. Returns a job id immediately; the media does not exist yet. Poll get_job after about eta_seconds to collect the result. Read get_model first unless the input fields are already known. A reference image is passed as a public https URL, or as a base64 data URI for a local file.
get_job	The state of one generation. While it runs, reports progress and how long it has left. Once it finishes, returns the output URLs and what the run cost. Also returns the finished image itself, so the result can be judged rather than guessed at.
list_jobs	Recent generations on this account, newest first. Use it to find a job id from an earlier session, or to see what is still running.
upload_media	Put a file on the Visual Sandbox CDN and return its URL, so it can be passed to generate as a reference image. Takes a base64 data URI for a local file. A file already on a public https URL needs no upload — pass that URL to generate directly.
list_presets	Saved recipes: a model plus the inputs and reference images that reproduce a particular look. Run one by reading it with get_preset and passing its inputs to generate.
get_preset	One saved recipe in full: its model, every input it fixes, and the fields it leaves for the caller to fill in.
get_balance	Credit left on this account, in US dollars, and the markup its prices carry. Check it before a run that costs real money, and when a generate call is refused for want of credit.
top_up	A checkout link for adding credit to this account. Returns a URL for the person to open; it charges nothing by itself and no card is touched until they complete the page. Call it when generate is refused for want of credit, or when get_balance is low and they ask to add more. Give them the link — never try to complete a payment.
list_skills	The Visual Sandbox skill packs: written guidance on how to prompt each family of model well. Read the one that fits before writing a prompt. Prompt craft changes the result far more than the choice of model does.
get_skill	Read one skill pack. Call this before writing your first prompt for a family of model: 'vsb-image-prompting' for any image, 'vsb-nano-banana' for the Nano Banana models, 'vsb-ugc-people' for a person in an advert, 'vsb-video' or 'vsb-p-video-prompting' for video, 'vsb-audio' for sound.