AI Content Drop Plugin for AI Agents

OAuth sign-in has been verified. Follow the connection steps below.

Give your coding agent a video and image studio. Ask for a product ad in the chat you are already in; the agent picks a model, quotes the credit cost, runs the generation and polls it to done. No copy-paste between tools, no separate dashboard.

Connection availability is described in the readiness notice above. Check the deployed server's tool list before relying on a workflow.

Quick install

Pick your agent. Each route ends with the same thing: an MCP connection to https://aicontentdrop.com/mcp.

Agent	Guide
Claude.ai and Claude Desktop	Claude.ai and Claude Desktop guide
Claude Code	Claude Code guide
Cursor	Cursor guide
Codex	Codex guide
VS Code	VS Code guide
Kimi Code	Kimi Code guide
Perplexity	Perplexity guide
Anything else that speaks MCP	MCP server reference

The whole configuration, for any client that reads the standard format:


{
  "mcpServers": {
    "aicontentdrop": {
      "type": "streamable-http",
      "url": "https://aicontentdrop.com/mcp"
    }
  }
}

Browsing costs nothing and needs no credential. Generating needs a sign-in: OAuth in hosts that support it, or an API key. See Authorizing.

What your agent can do

42 tools over MCP. The read tools answer with no credential at all, so an agent can research the catalogue and quote a price before anyone signs anything.

Tool	Credential	What it does
`list_models`	none	Every video or image model with its flat credit cost.
`estimate_credit_cost`	none	What a generation costs before you run it, as a signed `quote_id`.
`search_articles`	none	Search the published guides and model comparisons.
`get_article`	none	Read one guide in full.
`generate_video`	sign-in	Start a video generation from a `quote_id` and return a job id.
`generate_image`	sign-in	Start an image generation from a `quote_id` and return a job id.
`get_generation`	sign-in	Poll a video or image job to completion.
`list_generations`	sign-in	Recent generations on the account.
`get_account`	sign-in	Plan name and credits remaining.

Where the studio surface is enabled, the same server also lists the studio tools — a finished ad, a UGC presenter video or an avatar from a brief — under the same quote contract. They are documented in the MCP reference and walked step by step by the marketing-studio-ad skill.

Where the board surface is enabled, it also lists the board and research tools: the agent lays ideas, assets and draft generations out on a Visual Canvas board you can open in the browser, renders a card only with a quote, and can search the public ad libraries, save an ad, decode it into a brief and place it on the board. Results are labelled demo data whenever the live ad sources are not configured. The canvas-board skill walks the board flow.

Every generation starts with a quote: estimate_credit_cost returns a quote_id that is valid for 15 minutes and bound to the model, duration, resolution and quantity, and generate_video and generate_image refuse to run without one. A retry with the same quote_id returns the original job rather than a second charge.

Generation is submit-and-poll, never blocking: generate_video hands back a job id immediately and get_generation reports on it. A render takes minutes, which is longer than any single tool call should hold a connection open.

Two servers

https://aicontentdrop.com/mcp — the product server, all 42 tools.
https://aicontentdrop.com/mcp/docs — documentation only, the two article tools and nothing

that can spend credits. Useful when you want a retriever wired into a shared workspace without handing it the ability to bill anyone.

In the browser, with no install at all

The same tools exist as in-page tools on document.modelContext, the W3C WebMCP draft. A browser agent standing on https://aicontentdrop.com sees them and acts as the signed-in visitor, with no API key and no configuration — the page already knows who you are.

That surface carries a different set of tools, because a signed-in session can do things an API key cannot: read your credit balance, list your billing history, and check your own generations. See the MCP server reference for the full list and the annotations each tool carries.

The models

62 video models and 29 image models behind one credit system, including Kling 3.0, Veo 3, Sora 2, Seedance, Hailuo, Wan, Nano Banana Pro and GPT Image 2. One balance covers all of them; there is no per-model subscription to manage.

list_models is the authoritative list at any moment — it is generated from the same table the account is billed against, so it cannot drift from what you are charged. It also only ever advertises models the generation tools can actually run.

Three steps

Connect — add the MCP server with one of the guides above.
Authorize — sign in through OAuth where the host supports it, or

create an API key and give it to your agent, once.

Ask — "make me a 9:16 product ad for this photo" and let the agent pick

the model, quote the cost, and run it.

Authorizing

Claude, Perplexity and Kimi Code sign you in through OAuth: the host opens aicontentdrop.com, you approve a consent screen, and there is no key to paste. Every other client takes a key. Create one at Settings → Integrations and pass it as Authorization: Bearer acd_live_…. Keys are shown once.

An agent with no key is not stuck: it can read the catalogue, price a job, and tell you exactly what a key would let it do. The full authorization contract, including the machine-readable discovery documents, is at the authorization contract, /auth.md.

Who can use it

The plugin, the MCP tools and the skills are part of every paid plan. The free tier can read the catalogue and pricing, which is deliberate — an agent should be able to research the product and quote a price before anyone pays.

Agent-driven generations always bill credits at the standard rate. They do not draw on the included-model allowance that paid plans get in the web app: an automation loop is exactly the workload that allowance is not priced for, and metering it honestly is what keeps it available to everyone else. Prices are on the pricing page.

Command line

The same API, without an agent:


npm install -g aicontentdrop
acd models --type video
acd cost kling_3_0
acd generate "a cinematic product shot of a matcha tin" --model kling_3_0
acd status <id>

acd models and acd cost work with no credential.

Frequently asked

Do I need an API key to try it? No. The read tools, the whole catalogue and all pricing are open with no credential.

What does a generation cost? It depends on the model, from single digits to the high tens of credits. Ask estimate_credit_cost for the exact number before you run anything. It returns a quote_id that the generation call requires, and the call is refused if the price or the parameters no longer match the quote.

What happens if a generation fails? Nothing is charged. Credits are deducted on success only, so a failed or timed-out render costs you nothing and there is no refund to chase.

Can the agent spend my money without asking? The tools that spend are marked readOnlyHint: false, which is the signal your agent host uses to ask you first. generate_video and generate_image are the only two that spend credits, and neither runs without a quote_id from estimate_credit_cost, so the price is always on the table before the call. The server re-checks the quote itself, so a host that skips the confirmation dialog gets the same protection.

Is my prompt or output sent anywhere else? Generations run on the platform and the results are stored against your account. The in-page tools return links to your own media only when you explicitly ask for them, so a listing does not push playable URLs into an agent transcript by default.

Which MCP protocol revision? Streamable HTTP negotiating 2025-06-18, with the older HTTP+SSE transport accepted for compatibility. Both are live; see the MCP server reference.

Where are the skills? The skills catalogue — installable craft guides that teach an agent how to pick a model, write a prompt that works, and read an API failure correctly.