About Campaign Pack
Campaign Pack is a hosted MCP server for producing an advertising campaign from a product photo and a brief. It is for marketers and teams preparing client-facing campaign materials, with processing on Campaign Pack’s own GPUs and no third-party model provider in the product path.
Key features
- 01Create campaign plans, ideas, scripts, and creator briefs
- 02Generate product stills from an uploaded photo
- 03Produce a client-ready PDF
- 04Track campaign progress and queue position
- 05Share live runs and handle approval requests
Use cases
- 01Build an ad campaign from a product description and photo
- 02Prepare scripts and a creator brief for a launch
- 03Share an in-progress campaign with a teammate
- 04Review and approve an outbound action during a run
MCP support
The hosted MCP server lets clients start campaigns, check progress, retrieve finished packs, upload photos, share access, handle approvals, and abort runs. It connects at https://tools.campaignpack.ai/mcp and requires a Campaign Pack account.
Configuration
Remote servers need no local install. Pick your client, then copy the config or install in one click.
https://tools.campaignpack.ai/mcpclaude mcp add --transport http ai-campaignpack-campaignpack https://tools.campaignpack.ai/mcp{"mcpServers":{"ai-campaignpack-campaignpack":{"type":"http","url":"https://tools.campaignpack.ai/mcp"}}}Open in Cursor{"servers":{"ai-campaignpack-campaignpack":{"type":"http","url":"https://tools.campaignpack.ai/mcp"}}}Open in VS Code{"mcpServers":{"ai-campaignpack-campaignpack":{"type":"http","url":"https://tools.campaignpack.ai/mcp"}}}{"mcpServers":{"ai-campaignpack-campaignpack":{"serverUrl":"https://tools.campaignpack.ai/mcp"}}}Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.
Tools
Details
From the project READMECampaign Pack MCP server
Turn a product photo and a couple of sentences into a finished ad campaign — without leaving Claude, Claude Code or Cursor.
You get a campaign plan, campaign ideas, ad and social scripts, a creator brief, product stills, and a client-ready PDF you could send to a customer.
Everything runs on Campaign Pack's own GPUs. No third-party model provider is in the path, which is why an unreleased product photo can go into it.
Hosted server — there is nothing to install or self-host. This repository is the connection details and the tool reference.
Connect
Claude Desktop / Claude.ai
Settings → Connectors → Add custom connector, and give it:
https://tools.campaignpack.ai/mcp
A browser window asks you to sign in to Campaign Pack and confirm. That is all — no token to copy.
Claude Code
claude mcp add --transport http campaignpack https://tools.campaignpack.ai/mcp
Cursor
~/.cursor/mcp.json:
{
"mcpServers": {
"campaignpack": {
"url": "https://tools.campaignpack.ai/mcp",
"headers": { "Authorization": "Bearer YOUR_TOKEN" }
}
}
}
Get a token from the account panel at tools.campaignpack.ai → 🔌 Connect → Generate token. Name it after the machine you are putting it on, so you can disconnect that one later without touching the others.
You need a Campaign Pack account either way. Sign up free.
How to use it
Describe the product to your assistant the way you would to a person, then ask for a campaign. Your assistant writes the brief — that is the point of this being an MCP server rather than a form. The more specific the brief, the better the pack.
I sell a stainless steel water bottle, direct to consumers, keeps drinks cold for 24 hours. The buyer is a city commuter who already owns a bottle and wants a better one. Here's the product photo — make me a campaign.
Runs take minutes, not seconds. The assistant starts the job, gets an id back, and polls. You can watch the same run in a browser while it works.
Tools
| Tool | What it does |
|---|---|
start_campaign | Start a pack. Returns a job id and a URL to watch it |
check_campaign | Status, recent steps, queue position |
get_pack | The finished folder, the written overview, the PDF |
upload_product_photo | Put a real product photo in your workspace first |
share_campaign | Give a teammate access while the run is still going |
list_campaign_access | Who can see a campaign |
unshare_campaign | Take access back |
answer_approval | Approve or refuse anything the run wants to send outward |
abort_campaign | Stop a run |
Sizes
| Size | Time | What you get |
|---|---|---|
text | ~1 min | Plan, ideas, scripts, creator brief, PDF |
lite | ~10 min | The above plus two product stills |
The image size edits your photo rather than inventing a product, so upload a real one. Without a photo it refuses, which is deliberate: generated-from-nothing product images are not usable in an ad.
Sharing a live run
A campaign you start from your own assistant can be followed by a teammate from theirs, while it is still running. From their own account they can watch its progress, answer the approval prompts it raises, stop it, and read the finished pack. It is the same campaign in the same folder, not a copy.
Continuing a finished campaign with another turn is a browser job today — the tools here start a campaign and follow it, they do not add turns to one that has already run.
share_campaign(session_id, "[email protected]")
They need their own Campaign Pack account — sharing grants access, it does not create one.
Privacy
- No third-party AI API is in the product path. Not OpenAI, not Anthropic, not Midjourney, not ElevenLabs. The models run on hardware Campaign Pack owns.
- A run that wants to send your data anywhere outside pauses and asks a human
first, through
answer_approval. - Your assistant never sees your Campaign Pack password. Claude connects by OAuth; Cursor uses a token you issue and can revoke.
- Disconnect any client from the account panel at any time.
Limits worth knowing
- Runs serialise. There is one GPU, so one campaign per account at a time.
- A finished pack stays in your workspace; download the PDF from the website.
Links
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.