Skip to main content
Setup happens once. Connect the server, connect your own tools, then run onboarding, which records where your work lives and the rules your commands follow.

Before you begin

  • A Snow Doughnut subscription. Subscribing at snowdoughnut.com sends a welcome email with your personal server URL.
  • An AI that supports MCP servers: Claude, ChatGPT, Cursor, or another MCP-compatible agent.
  • The tools your marketing already lives in: where documents are kept, where work is managed, and the platforms that execute, such as an ads manager MCP, your email platform, and your CMS.

1. Add the server to your AI

Add the server URL from your welcome email as a connector or MCP server in your AI’s settings. Most agents take a remote server entry shaped like this:
Where your agent offers a connectors screen instead of a config file, paste the same URL there. Restart or reload the agent so it reads the new server.

2. Connect your own tools

The commands read and write your work through the tools you connect alongside the MCP, in the same AI:
  • A docs home, where filled workbooks, briefs, and libraries live. Notion, Google Drive, Confluence, or whatever your team already uses.
  • A work home, where tasks and projects are managed. ClickUp, Asana, Notion, or any tool with a connector.
  • Execution tools, which do the building and sending: an ads manager MCP for campaigns, your email platform, your CMS.
The MCP is tool-agnostic. It records which tools you chose and points every command at them; it does not require any particular one.

3. Run onboarding

Send one message:
Set up Snow Doughnut
Your AI calls setup_snowdoughnut. The first question is whether you are starting fresh or joining a team.

Joining a team that’s already set up

Say “I’m joining a team that’s already set up” and point to where the team’s setup lives: a Google Drive folder, a Notion page, or wherever it was saved. The MCP reads the team’s setup record there, confirms what it found in one line, and you are ready. Nothing the team already answered is asked again.

A new setup

  1. Save location first. The first question is where to save. Nothing else runs until there is a place to write to, because every answer after it gets saved there.
  2. Recommended or advanced. Recommended applies the framework’s full defaults in one step: every channel, placement, ad type, objective, creative type, and naming pattern in the dictionaries, and the default reporting schedule. You review the summary and change anything you want. Advanced walks every section and asks as it goes.
  3. The sections, in order, each opening with a one-line explanation of what it sets up and why, and a link to the page behind it:
    1. Locations. Where each kind of work lives. Each can be a different tool: briefs in Google Drive, creative in Dropbox, brand files in iCloud. See the asset directory.
    2. Dictionaries. Your naming conventions and UTM standards. Where a value carries a number, the dictionary records a pattern, such as OFFXX for a percent-off offer, rather than every amount.
    3. The schedule register. The reporting cadence: when each report runs, who approves it, whether it auto-sends, and where it lands.
    4. Standing rules and defaults. Standing rules and creator defaults per deliverable type.
You don’t have to be all new or all existing. Point any section at documents you already have, and the wizard reads them instead of asking.

The setup record

Your answers are saved as a setup record at the location you chose in step 1. The record holds where each kind of work lives, your dictionaries, your schedule, and how far through setup you are. The MCP remembers only where the record is. That is what makes setup resumable and shareable. Stop at any point, and “Continue Snow Doughnut setup” reads the record and picks up at the next unanswered step. A teammate who points to the same record joins your setup. Stages 1 to 6 of the setup path are the minimum before your first campaign; the rest can follow.

What setup does not ask

Who owns a folder, who can open a tool, or who adds teammates to it. Permissions are managed in your own tools. Setup records where things live, not who may open them. See what this framework covers.

4. Try a command

Ask for something real:
Name my new Meta prospecting campaign for Glacier Boots
The answer comes from your own dictionaries. If a value is missing, the AI asks you and records your answer, which is the missing-value rule working as designed. The full set is on operation commands.
Type: setup
Domains: mcp, setup
Provides: mcp-onboarding
Uses: creator-defaults, flat-file, glossary, naming-convention, reporting-cadence, utm-rules
Triggers: “Set up Snow Doughnut” · “Continue Snow Doughnut setup” · “I’m joining a team that’s already set up”Asks, one per turn, in order:
  1. New setup, or joining a team that’s already set up? Options: new, joining. Skip when: a setup record is already linked.
  2. Joining: where is the team’s setup? Options: your answer (a link or location). Then read the setup record, confirm in one line, and stop.
  3. New: where should Snow Doughnut save your setup? Options: your connected docs tools. Nothing else runs until this is answered.
  4. Recommended setup, or advanced? Options: recommended (full defaults, review at the end), advanced (every section). Recommended first.
  5. Advanced, per section: locations per kind of work, then dictionaries, then the schedule register, then standing rules and creator defaults. Options: asset-locations, naming-dictionary, reporting-cadence, creator-defaults. Skip when: recorded, or an existing document answers it.
Output: A setup record saved at the user’s chosen location: locations per kind of work, dictionaries, schedule register, defaults, and progress. The MCP stores only where the record lives.
Last modified on October 30, 2026