MCP reference

The Print.App MCP server for developers: connecting every client and API, the tools, and how signing in works.

Just want to connect? Follow MCP server, or hand your agent the Quickstart for agents.

Print.App runs a remote Model Context Protocol server, so AI agents such as ChatGPT, Claude, Cursor or your own can work in your account through the REST API: build a design from a brief, look at it and fix it, then put it on a product.

https://api.print.app/mcp

A client connects one of two ways, both with a secret key of your store:

• Sign in (OAuth): apps such as ChatGPT and Claude open a Print.App page where you paste the key and allow access. The app never sees the key: it gets its own access, tied to that key.
• Send the key: developer tools and your own code send Authorization: Bearer sk_… with every request.

An agent connected with your key can change and delete anything in your account that the API can. Create a separate key for each app or agent, labelled so you recognise it. Revoking the key in the admin disconnects it, including apps that signed in with it.

Connect a client

ChatGPT

1. In ChatGPT, add a custom MCP server (it is in developer mode; see OpenAI’s Connecting custom MCP servers in ChatGPT for where it is on your plan).
2. Name it Print.App, enter the URL https://api.print.app/mcp, and choose OAuth as the authentication.
3. ChatGPT opens the Print.App connect page. Paste a secret key of your store and click Allow access.

Claude

In Claude (claude.ai or the desktop app), add a custom connector with the URL https://api.print.app/mcp, then click Connect. Claude opens the Print.App connect page: paste a secret key of your store and click Allow access.

Claude Code

claude mcp add --transport http printapp https://api.print.app/mcp \
  --header "Authorization: Bearer sk_…"

Cursor

In .cursor/mcp.json (or your global Cursor MCP settings):

{
  "mcpServers": {
    "printapp": {
      "url": "https://api.print.app/mcp",
      "headers": { "Authorization": "Bearer sk_…" }
    }
  }
}

VS Code

In .vscode/mcp.json:

{
  "servers": {
    "printapp": {
      "type": "http",
      "url": "https://api.print.app/mcp",
      "headers": { "Authorization": "Bearer sk_…" }
    }
  }
}

Claude API

With the MCP connector of the Messages API:

{
  "mcp_servers": [{
    "type": "url",
    "url": "https://api.print.app/mcp",
    "name": "printapp",
    "authorization_token": "sk_…"
  }]
}

Codex

In ~/.codex/config.toml, with the key in an environment variable rather than in the file:

[mcp_servers.printapp]
url = "https://api.print.app/mcp"
bearer_token_env_var = "PRINTAPP_SECRET_KEY"

export PRINTAPP_SECRET_KEY=sk_…

OpenAI API

With the remote MCP tool of the Responses API:

{
  "tools": [{
    "type": "mcp",
    "server_label": "printapp",
    "server_url": "https://api.print.app/mcp",
    "authorization": "sk_…",
    "require_approval": "never"
  }]
}

Send authorization with every request: OpenAI doesn’t store it. Leave out require_approval to approve each tool call yourself.

Any MCP client that speaks Streamable HTTP works the same way, sending a key or signing in. The server is stateless: it answers JSON over POST and keeps no sessions.

How signing in works

Clients find the sign-in on their own: a request without credentials gets a 401 that points to https://api.print.app/.well-known/oauth-protected-resource/mcp, and from there to the authorization server’s metadata at https://api.print.app/.well-known/oauth-authorization-server. It supports dynamic client registration and client ID metadata documents, PKCE (S256) and refresh tokens. Access lasts an hour and the client renews it on its own; it stops within a minute of the key it was approved with being revoked.

Redirect URIs must be https addresses, as ChatGPT’s and Claude’s are. Apps that sign in through a local address (http://localhost:…) can’t register; send the key as a header from those instead.

What an agent can do

Area	Tools
Account	get_account
Designs	list_designs, get_design, create_design, update_design, duplicate_design, delete_design, manage_pages
Page content	get_design_content, edit_design, design_ops_reference, preview_design
Design folders	list_design_folders, manage_design_folder
Files	list_files, get_file, upload_file, manage_file
Graphic folders	list_graphic_folders, manage_graphic_folder
Fonts	list_fonts, manage_font
Palettes	list_palettes, manage_palette
Products	list_products, get_product, set_product_designs
Connectors	inspect_product_page, test_connector, list_connectors, manage_connector
Profiles	list_profiles, manage_profile (resource and PDF profiles)
Settings	get_settings, update_settings
Projects	list_projects, get_project, manage_project
Editor links	create_editor_link, manage_editor_link

Each tool calls the REST API with your key, so the same rules, limits and errors apply. Tools that only read are marked read-only; tools that can delete are marked destructive, so clients that ask before running those will ask.

update_design also sets what a design uses (its connectors, resource profiles and PDF profile) and its own settings over the store’s.

The server also offers one prompt, set_up_connector (Set up a connector), for clients that list MCP prompts: it takes a product page and a goal, and writes out the steps below for the agent.

How an agent builds a design

1. create_design at the trim size, with bleed and a safe margin.
2. upload_file for logos, photos and SVG graphics.
3. get_design_content to read the page size and safe area, then edit_design with a batch of operations: text, shapes, graphics, photo frames, QR codes, calendars. The batch saves as one change, or not at all.
4. preview_design returns images of the pages, so the agent can look at its work and fix what it sees with another edit_design.
5. set_product_designs to sell it on a store product, or create_editor_link to share it.

Try it with a prompt like “Create a two-sided 85 × 55 mm business card for a bakery called Crumb & Co with 3 mm bleed: logo from https://…/logo.png, the name, job title and phone as fields customers fill in, and a warm palette. Show me both sides.”

How an agent sets up a connector

1. inspect_product_page reads the product page of your store: each option control with a selector that finds it and the choices it sends, the Print Options fields with their roles, and the designs the product opens.
2. get_design and preview_design show the design’s pages.
3. test_connector plays the connector, before it is saved, against the design for every choice: what the page sends, which mapping matches, what the editor does, with images of the pages each choice shows. It reports what stands in the way: a selector that finds nothing, choices without a mapping, pages the design doesn’t have, a design that isn’t on the product, another connector fighting this one.
4. manage_connector saves it and attaches it to the designs (create with designIds), and test_connector runs once more with its id.

The walk-through for store owners is Set up connectors with AI.

Things to know

• Page content can only be edited on accounts that use the new editor. Others get editor_not_supported.
• SVG graphics go in as files. The firewall in front of the API refuses requests that carry markup, so an agent should upload an SVG with upload_file and place it as an asset, not write SVG into an operation. It stays vector in print. A tool call with SVG or HTML in its arguments fails with an HTTP 403 before it reaches the server, so the client reports a connection error rather than a tool error.
• Fonts must be families your account has. list_fonts lists them; manage_font imports any Google font.
• Previews are redrawn each time a design is saved. They are for checking a layout; the PDF is made from the design itself.
• Settings that run code in your store (custom JS and CSS), send data elsewhere (webhook addresses) or delete customers’ projects can’t be changed through the MCP or the API, and the editor in use is read-only: change those in the dashboard. update_settings is marked destructive, so clients ask before running it.
• Product pages are read as your server sends them, and only pages of your own store. Options a theme builds in JavaScript after the page loads aren’t seen; a store behind a password answers with its password page (give test_connector the choices as values instead).