Your ad stack, inside Claude.
We speak the Model Context Protocol. Point your assistant at our server, paste in an API key, and it can generate ad creative and run audience simulations directly in your workspace — no tab switching, no copy-paste, no glue code.
{
"mcpServers": {
"goosecabbage-ads": {
"type": "http",
"url": "https://mcp.goosecabbage.com/mcp",
"headers": {
"Authorization": "Bearer gc_live_…"
}
}
}
}What your assistant can actually do.
Each one is a real endpoint in your workspace — the same code path the app uses, with the same permissions, credits, and safety checks.
generate_ad_imageクリエイティブGenerate an on-brand ad image from a prompt — composition, lighting, locked headlines, and CTA text. Returns a hosted asset.
generate_ad_videoクリエイティブRender a short ad video in 16:9, 9:16, or 1:1, optionally from a first frame or reference image.
run_simulationシミュレーションStart a pre-flight audience simulation against a universe — a whole synthetic focus group, before you spend anything.
get_simulation_statusシミュレーションCheck progress and cost while a simulation runs, so the assistant knows when to look at results.
get_simulation_resultsシミュレーションRead per-concept scores and the aggregate audience response once the run completes.
list_simulationsシミュレーションList recent simulations in the workspace, so you can pick up where you left off.
Three steps, about two minutes.
Works with Claude Desktop, Claude Code, and any other MCP-capable client.
Create an API key
gc_live_ and is shown once. Give your assistant its own key so you can revoke it without touching anything else.Add the server
Just ask
An agent with your key, not a blank cheque.
Giving an assistant access to an ad account should not mean giving up control of it. It doesn't here.
It acts as you, nowhere else
Every call carries the key from your own config. Our server holds no credential of its own, so it can only ever reach the one workspace that key belongs to.
The same rules as the app
Tools call the same REST endpoints the product does, so credit accounting, rate limits, content checks, and approval gates all apply exactly as they would if you'd clicked the button.
Nothing goes live by accident
These tools generate creative and run simulations. Publishing spend still goes through your normal approval flow — see Governance & Approvals.