Using images with an AI agent

A connected agent can use images in a design by referring to workspace assets. The asset tools appear only when the connection and workspace allow them. Start with connecting an agent, then use the tools reference for the current tool names and access rules.

Bring in an image#

For a public HTTP(S) image, call import-asset. It downloads the image into durable workspace storage and returns an assetId that can be used in a Design Document. For a local file, call prepare-asset-upload with its filename, byte count, and SHA-256. Transfer the bytes through the scoped HTTP upload it returns, then use the resulting asset reference. Do not put a large base64 image in an MCP message or a Design Document.

Use list-assets to browse assets you can access and get-asset to inspect one. Then refer to its assetId when creating or patching an image element. Validate the document before writing it, and render the saved result to confirm the image loads.

Repair an older design#

repair-design-assets can replace retrievable external or embedded image sources with durable workspace assets. It works on selected image element keys in an existing design and requires write access. A browser-only blob: URL cannot be fetched later; upload a replacement from the browser that still has the original file. Check the result in the editor and keep a version snapshot before larger edits.