Is there an MCP server for AI video and image generation?

Yes. TurboMax AI exposes a remote MCP server at https://turbomaxai.com/mcp (Streamable HTTP, JSON-RPC 2.0). An MCP client can list the available image, video and audio models with their live credit prices, start a generation, follow it to completion and read the result URL.

Do I need an account or API key to try the TurboMax MCP server?

No key is needed to connect or to use the discovery tools: initialize, tools/list, list_models, get_model, estimate_cost, list_chat_models and browse_gallery work anonymously. Checking a wallet needs a read-only key, which is free. Spending credits needs a generate key, which requires a paid account.

How does an AI agent pay for generations through MCP?

It spends credits from the key owner's prepaid wallet. There is no per-call billing relationship with the agent itself. Each generate_media call can carry a maxCredits ceiling; if the real cost would exceed it, the server refuses before anything is charged.

Can an agent with a TurboMax key change my account or billing?

No. API keys are fenced to a short allowlist of routes. They can never reach account settings, billing, administration, publishing to the public gallery or deleting history, whatever the key's scope.

Which MCP clients work with it?

Any client that supports remote MCP servers over HTTP with a custom Authorization header, including Claude Code, Cursor and VS Code. Clients that only launch local processes can connect through a remote-to-stdio bridge such as mcp-remote.

MCP server for AI video, image and music generation: what TurboMax offers

TurboMax AI runs a public Model Context Protocol (MCP) server so Claude, Cursor, VS Code and your own agents can generate video, images and music, chat, and read a prepaid wallet. What it exposes, how it is billed and how it is kept safe.

The Model Context Protocol (MCP) is an open standard that lets an AI assistant call external tools through one consistent interface. TurboMax AI publishes an MCP server, so an assistant or an autonomous agent can do what a person does in the TurboMax studio: pick a model, check the price, generate a video, image or piece of music, and collect the result.

Endpoint: https://turbomaxai.com/mcp — Streamable HTTP, JSON-RPC 2.0, stateless.

What an agent can do

The server exposes 13 tools. Which ones cost money is stated in each tool's description, so an agent can decide before it calls.

Tool	Purpose	Spends credits?
`list_models`, `get_model`, `estimate_cost`	The live catalogue: tasks, aspect ratios, resolutions, duration range, and price in credits	No
`generate_media`	Start an image, video or audio job	Yes
`get_task`	Status, progress and result URL; can wait up to 55 seconds	No
`cancel_task`	Stop a running job and release its hold	Refunds
`retry_task`	Re-run a failed or cancelled job	Yes
`upload_source_media`	Upload a source image, video or audio file for image-to-video and editing	No
`list_chat_models`, `chat`	Text models, billed per message	`chat` does
`get_wallet`, `list_generations`	Balance, daily limits, ledger and history	No
`browse_gallery`	The opt-in community gallery	No

Nothing about a model is hard-coded in the server. Model ids, parameter limits and prices are read from the same catalogue the website uses, so an agent that calls list_models always sees what is actually available today.

Billing: prepaid credits, not a subscription

TurboMax is pay-per-use. A person tops up a wallet (minimum $5, $1 = 200 credits) and credits are spent only when something is generated. There is no monthly subscription. An agent spends the key owner's credits, so the owner stays in control of the total exposure: they decide how much to put in the wallet, which scope a key gets, and when to revoke it.

Two details matter for automated callers:

The price you see is the price that is held. When a job starts, the estimated cost is reserved from the wallet. When the provider reports the real cost, the job settles to the actual amount and any difference is refunded.
Failures cost nothing. A job that fails, times out or is cancelled is refunded in full.

Keys and scopes

Access uses a personal API key beginning tmx_, created on the MCP page of the TurboMax app. The key is shown once; only a hash is stored. Two scopes exist:

Read only — browse models, check the wallet and history, poll a job. Free for any signed-in account.
Generate — everything above, plus upload, generate, cancel, retry and chat. It spends credits, so it requires an account that has made a paid top-up.

Keys expire (30 days, 90 days or one year), at most five can be active at once, and revoking one takes effect immediately.

Safety for unattended use

An agent that spends money with no human watching needs limits that do not depend on the agent behaving well. The server enforces them:

Spending ceiling per call. generate_media accepts maxCredits. If the cost would exceed it, the request is refused with cost_exceeds_max and nothing is charged.
Daily caps. The account's normal daily credit and generation limits apply to keys exactly as they do to the website.
Prompt screening. Prompts go through the same content policy as the website; a refused prompt costs nothing.
A narrow fence. A key cannot change the account, billing or settings, cannot publish work publicly and cannot delete history.
No invented prompts. The server never writes a prompt for the user. An empty prompt is refused rather than filled in.

Try it without a key

Discovery needs no credentials. This request lists the image models with their current prices:

curl -s https://turbomaxai.com/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"list_models","arguments":{"type":"image"}}}'

A machine-readable description of the server is published at https://turbomaxai.com/.well-known/mcp/server-card.json. For step-by-step client setup see How to connect Claude, Cursor and VS Code to the TurboMax MCP server, and for the agent workflow see How an AI agent generates video through MCP.