Connect an MCP client

Bannerify runs a Model Context Protocol (MCP) server at https://api.bannerify.co/mcp. This page is for anyone who works with an AI assistant, such as Claude, Cursor, or VS Code, and wants it to use their Bannerify templates. Once the server is connected, the assistant can:

List your templates and the layers you can change in each one.
Generate an image or a PDF from a template.
Compose a one-off image from shapes when no template fits.

Before you start

A Bannerify account. The template tools need at least one template, and composing an image from shapes needs none.
An MCP client that supports remote servers over HTTP and OAuth sign-in.

Add the server

The server URL is https://api.bannerify.co/mcp, and it uses the Streamable HTTP transport. The Agents & MCP page in the dashboard shows the same setup steps for each client, with buttons to copy the URL and the commands.

The Agents & MCP page with Claude, ChatGPT, Codex, Cursor, and Others, and the setup steps for Claude — The Agents & MCP page in the dashboard.

Examples

Open only what you need

Claude Code

Add the server, then start the sign-in from inside Claude Code:

claude mcp add --transport http bannerify https://api.bannerify.co/mcp

Run /mcp in Claude Code and follow the steps in your browser.

To share the server with your team, put a .mcp.json file at the root of your project:

{
  "mcpServers": {
    "bannerify": {
      "type": "http",
      "url": "https://api.bannerify.co/mcp"
    }
  }
}

Claude

Claude adds a remote server as a custom connector. On the web:

Open Customize > Connectors.
Click + Add, then Add custom connector.
Enter Bannerify as the name and https://api.bannerify.co/mcp as the remote MCP server URL, then click Continue.
Finish the prompts and click Add.

On a Team or Enterprise plan, an owner adds the connector first in Organization settings > Connectors.

The Claude desktop app gets its connectors from your Claude account. Do not add the server to claude_desktop_config.json: that file is for local servers only.

The Claude help article on custom connectors lists which plans can add one.

Cursor

Add the server to .cursor/mcp.json in your project, or to ~/.cursor/mcp.json for every project:

{
  "mcpServers": {
    "bannerify": {
      "url": "https://api.bannerify.co/mcp"
    }
  }
}

Cursor supports OAuth sign-in for remote servers.

VS Code

Add the server to .vscode/mcp.json in your project:

{
  "servers": {
    "bannerify": {
      "type": "http",
      "url": "https://api.bannerify.co/mcp"
    }
  }
}

Codex

Add the server, then sign in:

codex mcp add bannerify --url https://api.bannerify.co/mcp
codex mcp login bannerify

The second command opens the Bannerify sign-in page in your browser.

For another client, enter the same URL and choose the HTTP transport with OAuth sign-in.

The first time your client connects, it has no token, so Bannerify sends it to the sign-in page. Your client opens that page in a browser. After you sign in, the client keeps the token and sends it with each request. You create no API key and copy no token.

The exchange works in four steps:

The client sends a request to https://api.bannerify.co/mcp with no token. The server answers with HTTP 401 and a WWW-Authenticate header that points to its metadata at https://api.bannerify.co/.well-known/oauth-protected-resource/mcp.
The metadata names the sign-in service, and the client sends you there.
You sign in, and the client receives an access token.
The client sends the token as Authorization: Bearer <token> with each request. Bannerify checks the signature, the expiry, and that the token was issued for the MCP URL.

Detail	Value
Valid for	`https://api.bannerify.co/mcp` only. A token issued for another URL gets HTTP 401
Workspace	The team workspace in the token. A token without a team workspace gives your personal workspace
Invalid or expired token	HTTP 401 with the error `invalid_token`. The client signs in again
Token storage	Bannerify does not store the token. The server verifies it on each request

Ask for a file

Give the assistant a plain request:

List my Bannerify templates, with the layers of each one.

Create a PNG from my "Lifecycle email" template. Set the headline to "Welcome back, An" and the qr layer to https://example.com/restart.

Create a PDF from my invoice template for Acme Studio.

Design a 1200 by 630 banner with a dark background and the text "Launch week", then show me the image.

Which plan is my Bannerify workspace on?

The assistant picks the tools it needs from Generate files with MCP tools, renders the file, and shows you the result.

Usage and limits

A generation from an MCP client counts toward the same monthly and daily limits as the REST API and the SDKs. The plan of the connected workspace applies.

Tool	Counts as a generation
`create_image`, `create_pdf`, `create_image_from_shapes`	Yes
`list_templates`, `get_template`, `get_project`	No

A render that fails does not count. A team workspace on the Free plan cannot generate files, and a new team workspace starts with a 14-day Automate 3k trial. The MCP server has no request rate limit. Quota and limits has the limit of each plan.

Disconnect

Remove the server from your MCP client, or remove the connector in Claude. The client stops sending its token. Bannerify keeps no copy of the token, so the token stops working when it expires.

Troubleshoot

Symptom	Fix
The client reports HTTP 401, or asks you to sign in again	The token is missing or has expired. Sign in again from your client
The browser does not open a sign-in page	Use a client that supports remote MCP servers over HTTP with OAuth sign-in
The Claude app does not show Bannerify	Add it as a custom connector in Customize > Connectors, not in `claude_desktop_config.json`
The assistant sees no templates	Ask it to run `get_project`. If the name shows a different workspace than the one with your templates, remove the server and connect again
`NOT_FOUND` for a template	Ask the assistant to run `list_templates` and use an id from the result. The tools see only the templates of the connected workspace
`USAGE_EXCEEDED`	The workspace reached its limit. See Quota and limits