An AI agent makes a design with Polotno in two steps. It writes a design as Polotno JSON, helped by the official skills and the published JSON Schema. Then a renderer turns that JSON into a PNG, JPEG, PDF, SVG, HTML, PPTX, GIF or MP4 file: the Cloud Render API, or one of two Node.js packages on your own servers. The design stays editable, so a person can open the same JSON in the Polotno editor and change it.
Step 1: the agent writes Polotno JSON
A Polotno design is a readable JSON document: a size, a list of pages, and typed elements on each page. An agent can write one, read it back, change one field, and check the result.
The polotno-design skill teaches an agent to produce designs: posters, social posts, multi-page decks and video, with visual proofs and variants from data. A second skill, polotno-sdk, covers integrating the editor. Both ship as a plugin for Claude Code and for Codex, and a cross-agent installer covers Cursor, Copilot, Windsurf, Cline and others. See Build with AI.
The JSON Schema is published. Give it to a model as context, then validate what comes back before you render it.
A coding agent can also search the current documentation over MCP, so it does not write against an API it remembers from training.
One line starts an agent on the design skill:
Read https://polotno.com/skills/design.md and follow it.Step 2: a renderer draws it
Three renderers take the same JSON. Pick by output format and by where the render has to run.
| Renderer | Where it runs | Output | Use it when |
|---|---|---|---|
| Cloud Render API | Polotno's cloud | PNG, JPEG, PDF, SVG, HTML, PPTX, GIF, MP4 | You want no servers to run. The agent sends JSON over HTTP and gets the file back. |
@polotno/image-export | Your Node.js process | PNG, JPEG | You need still images on your own servers with no headless browser. It does not draw video frames or Lottie animations. |
polotno-node | Your servers, with headless Chromium | PNG, JPEG, PDF, GIF, MP4 | You need PDF, GIF or MP4 on your own servers, or the design holds video or Lottie elements. |
@polotno/image-export runs in Node.js 22.13 or newer and needs network access, because it downloads the design's images and fonts. In Polotno's measurements it renders photo templates 2–5× faster than polotno-node and simple layouts about 15× faster, with a peak of 280 MB of memory against 700 MB.
Vector PDF has its own package, @polotno/pdf-export, which also runs in Node.js with no headless browser.
A minimal path
The agent writes the design as a JSON file. Your code validates it against the schema, then passes it to a renderer, which writes the image. The schema docs show the validation call and the image export docs show the render call.
For the same result with no renderer installed, send the JSON to the Cloud Render API. The Cloud Render API docs show the request.
MCP servers
Polotno SDK has no MCP server that renders designs. Its documentation MCP server searches the docs and the community forum, and it does not create or render anything. An agent renders by running one of the packages above or by calling the Cloud Render API over HTTP.
Polotno App, the desktop design editor, is a separate product with its own built-in MCP server. A connected agent creates, edits and exports designs in the app on your computer, while you work on the same files.
Polotno does not store designs for you. The JSON lives wherever the agent or your application saves it.
Who uses it
- Agents that return images – a marketing or sales agent answers with a finished graphic.
- Prompt-to-design in a SaaS – the model writes a first draft and the user finishes it in the embedded editor.
- Batch generation – an agent fills one template from rows of data and renders each variant.
- Deck tools – an agent reads an uploaded deck through PPTX import, edits the JSON, and exports the slides again.
FAQs
Can an AI agent generate images with Polotno?
Yes. The agent writes Polotno JSON, and the Cloud Render API, @polotno/image-export or polotno-node renders it to an image file.
Is there a Polotno MCP server that renders images?
Not in Polotno SDK. Its documentation MCP server only searches documentation, and rendering goes through the Cloud Render API or one of the Node.js packages. Polotno App, the desktop editor, has a built-in MCP server that exports designs from the app.
Does the agent need a headless browser?
Not for PNG, JPEG or vector PDF. @polotno/image-export and @polotno/pdf-export run in Node.js with no headless browser. GIF, MP4 and raster PDF on your own servers go through polotno-node, which runs headless Chromium. So does any design whose video or Lottie elements must appear in the output.
How do I stop a model from writing invalid JSON?
Give the model the published JSON Schema as context, validate the output, and send validation errors back to the model for a second attempt.
Can a person edit what the agent made?
Yes. The output is an ordinary Polotno design. Load it into the editor and every element is selectable and editable.
Can Polotno create the design from a prompt for me?
An experimental AI design generation endpoint returns a design from a text prompt. For decks, video and variants from data, use the polotno-design skill.
Does this need a Polotno licence?
The skills and the documentation MCP server are free to install. Rendering in production needs an API key, as the SDK does.
Related reading
- Build with AI: the skills, plugins and documentation MCP server.
- Build with AI docs: install commands.
- Rendering: cloud, self-hosted and browser rendering compared.
- Design schema: the JSON shape agents read and write.
