Generate files with MCP tools

The Bannerify MCP server has six tools. The assistant picks them from your request, so you rarely call one by name. Every tool works only on the workspace of the connection. Connect an MCP client first if you have not.

Tools

Tool	What it does	Counts as a generation	Arguments
list_templates	Lists the templates in the workspace. With includeLayers, it also lists the layers you can change	No	includeLayers (boolean, optional)
get_template	Reads one template and its layers	No	templateId (string)
get_project	Reads the workspace: its id, name, plan, and creation date	No	None
create_image	Renders a template and returns the image	Yes	templateId (string), modifications (object, optional), format (png, jpeg, or webp, default png)
create_pdf	Renders a template as a PDF and returns a link to it	Yes	templateId (string), modifications (object, optional)
create_image_from_shapes	Composes an image from shapes, with no template	Yes	canvas (object, optional), shapes (array), format (png, jpeg, or webp, default png)

Use a template id exactly as list_templates returns it.

Find a template

Ask for the templates and their layers:

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

The tool returns one entry for each template. Without includeLayers, an entry holds only the id and the name. With it, each layer comes with the name to use in modifications:

[
  {
    "id": "tpl_xxx",
    "name": "Lifecycle email",
    "layers": [
      { "name": "headline", "id": "shape:abc123", "type": "text", "suggestInput": "text" },
      { "name": "photo", "id": "shape:def456", "type": "image", "suggestInput": "src" },
      { "name": "qr", "id": "shape:ghi789", "type": "qrcode", "suggestInput": null }
    ]
  }
]

suggestInput names the field to set: text for a text layer, src for an image layer, and null for every other type. A layer without a name shows as Untitled.

Layer names come from the template editor, where you rename a layer by double-clicking it. Modifications explains how a name matches a layer.

Pass layer values

modifications is an object keyed by layer name. A string sets the text of the layer, and an object sets the fields of that layer. The array format of the REST API is not valid here.

Layer	Example value
Text	{ "headline": "Summer sale" }
Image	{ "photo": { "src": "https://example.com/photo.png" } }
QR code	{ "qr": { "qrcode": "https://example.com/restart" } }
Table	{ "prices": { "columns": ["Plan", "Price"], "rows": [{ "Plan": "Automate", "Price": "$29" }] } }
Any layer	{ "tag": { "visible": false } }

A full example for create_image:

{
  "templateId": "tpl_xxx",
  "format": "png",
  "modifications": {
    "headline": "Welcome back, An",
    "photo": { "src": "https://example.com/hero.png" },
    "qr": { "qrcode": "https://example.com/restart" }
  }
}

A layer that you leave out keeps the values from the template, and a name that matches no layer is ignored. Modifications lists the fields of every layer type.

Read the result

create_image returns the image inline, with a line that gives its format and size in bytes. Your client decides how to show or save the image.

create_pdf stores the file and returns a link:

Generated a PDF from template tpl_xxx. Download: https://images.bannerify.co/files/mcp/0b3c1c1e-5d6f-4c52-9e0a-8f6a4c0f2a11.pdf

The PDF has one page with the size of the canvas, and its text can be selected and searched. Generate PDFs has the details.

Compose an image without a template

Use create_image_from_shapes when no template fits. Give the tool a canvas and a list of shapes. It draws the shapes in order and returns the image, so the assistant can design a one-off banner or invoice.

The canvas object is optional. It is 1200 by 630 pixels and white unless you set these fields:

Field	Description
width, height	The size in pixels, from 16 to 4096 on each side
background	A solid color. The default is white
backgroundImage	An HTTPS URL, a data URL, or a CSS gradient such as linear-gradient(to right, #4f46e5, #9333ea). It wins over background
objectFit	How the background image fills the canvas: cover (default), contain, or none

Every shape has type, x, y, width, and height, in pixels from the top left of the canvas. The first shape is at the back and the last is on top. Three more fields work on every type: opacity (0 to 1), rotation (degrees, clockwise), and style.

Type	Required	Optional
text	text	fontSize (default 24), align (left, center, or right), lineClamp, color, fontFamily, fontWeight
frame	None	backgroundColor, borderColor, borderWidth, borderRadius
image	src	objectFit (fill, contain, cover, none, or scale-down; default cover), borderRadius
qrcode	qrcode	backgroundColor
barcode	barcode	None
icon	icon	color
star	None	star (1 to 5, default 1), color
table	columns, rows	footer, variant, stripe, compact, fontFamily, align (left, center, or right)

A few rules:

• icon takes an Iconoir name, for example check-circle or arrow-right. A name that Iconoir does not have fails the call.
• src takes an HTTPS URL or a data URL. A remote image loads with the same rules as in a template: PNG, JPEG, WebP, or SVG, and a failed download returns FETCH_IMAGE_ERROR.
• A table has one value per column in every row. This differs from modifications, where a row is an object. variant is line, grid (default), minimal, striped, compact, bordered, or primary-header.
• style takes CSS property names in camelCase, for example padding, boxShadow, lineHeight, and letterSpacing. A key that the renderer does not support fails the call with an Unrecognized key message. Set the size with width and height, not with style.
• A field on the shape, such as color or borderRadius, wins over the same key in style.
• A color is a hex value, rgb(), hsl(), or a CSS color name.

Limit	Value
Shapes	1 to 100
Canvas	16 to 4096 pixels on each side, and at most 12,000,000 pixels in total
Text	4,000 characters
Table	12 columns and 200 rows
src and backgroundImage	8,192 characters
qrcode value	2,000 characters
barcode value	200 characters

A full example, an invoice:

{
  "canvas": { "width": 800, "height": 560, "background": "#f8fafc" },
  "shapes": [
    { "type": "frame", "x": 0, "y": 0, "width": 800, "height": 120, "backgroundColor": "#0f172a" },
    { "type": "text", "x": 48, "y": 44, "width": 400, "height": 40, "text": "INVOICE", "fontSize": 30, "color": "#ffffff" },
    { "type": "text", "x": 500, "y": 52, "width": 252, "height": 24, "text": "INV-1042", "fontSize": 18, "align": "right", "color": "#94a3b8" },
    {
      "type": "table",
      "x": 48,
      "y": 176,
      "width": 704,
      "height": 180,
      "columns": ["Item", "Qty", "Total"],
      "rows": [["Logo pack", 1, "$29.00"], ["Banner credits", 500, "$19.00"]],
      "footer": ["", "Due", "$48.00"],
      "variant": "striped",
      "compact": true
    },
    { "type": "text", "x": 48, "y": 392, "width": 300, "height": 28, "text": "Thank you for your business.", "fontSize": 16, "color": "#475569" },
    { "type": "qrcode", "x": 632, "y": 384, "width": 120, "height": 120, "qrcode": "https://bannerify.co/invoices/1042" }
  ]
}

Bannerify checks the whole request before it renders, so an invalid request does not use a generation. A valid one counts like a generation from a template. See Quota and limits.

Handle errors

A failed call returns isError: true and a text message in one of two forms:

Form	Meaning
CODE: message, for example NOT_FOUND: template tpl_xxx not found	An error from Bannerify. The next table lists the codes
Invalid arguments. <field>: <message>	The arguments do not match the tool, and the message names the field. This form has no code

Code	Meaning	What to do
NOT_FOUND	The template does not exist in this workspace. A template of another workspace is not found either	Run list_templates and use an id from the result
USAGE_EXCEEDED	The workspace reached its monthly or daily generation limit	Wait for the next period, or see Quota and limits
BAD_REQUEST	A color or style value in create_image_from_shapes is not valid, or a value in modifications does not fit its layer field	Fix the value that the message names. For modifications, the message gives the position of the layer in your object, counted from 0, for example modifications[0].src
FETCH_IMAGE_ERROR	An image could not be downloaded, or is too large to decode	Use a public URL that returns an image
INTERNAL_SERVER_ERROR	The render failed, or an icon name does not exist in Iconoir	Try again, and check the icon names

A missing or expired token gets HTTP 401 with the error invalid_token before any tool runs, so a tool never returns UNAUTHORIZED. Your client then signs in again. RATE_LIMITED and TOO_MANY_REQUESTS do not occur, because the MCP server has no request rate limit. Handle errors lists the codes of the REST API.

Learn more

• Connect an MCP client
• Modifications
• Layer types