Canva

Use the installed canva CLI. Start with canva status. If it reports not_connected, run canva authorize-url and share only the returned connect_url. If the CLI reports outdated OAuth settings, have the user disconnect Canva in Settings and reconnect. Never request tokens in chat.

Run canva list-tools for live schemas and use only tools returned by that call. Treat omission as authoritative for the current task: never infer tool availability from this skill, a prior turn, a cached schema, or a tool expected to launch later. Follow returned input schemas exactly and include a concise user_intent. hatch_permission_overrides identifies argument-dependent permissions such as deleting pages. Some tools require Canva Pro, Enterprise, or available AI credits.

Save each raw response before parsing. Check result.isError, then read result.structuredContent when present or parse the JSON text block in result.content. Both response forms are valid. A local parsing failure does not mean a mutation failed: recover its response or inspect saved state before continuing. Never repeat a copy, creation, or commit just to obtain its output.

canva search-designs [--query <keywords>] [--continuation <token>]
canva get-design-pages --design-id <id>
canva list-tools
canva call-tool --name <tool-name> --arguments-json '<JSON object>'

Create designs and images

Use create-design for a new editable layout: social post, infographic, poster, flyer, presentation, document, or sheet. Put all necessary source facts and requested wording in brief; the tool does not inherit chat or connected-source context. Fetch the user-selected source first. For an exact size, state dimensions and orientation in the brief. If supplying format, include the orientation; an ambiguous format may resolve to a square. Pass outline only for a presentation outline the user supplied or approved.

Use generate-image for an explicitly requested standalone image, photo, illustration, artwork, or image of an infographic. Use one MEDIA reference per uploaded source image. This generates an image rather than an editable page layout. Use its returned media_id when editing a generated image again.

create-design cannot apply brand kits or brand templates. For an explicitly on-brand request, use the available legacy generate-design, get-design-candidates, and create-design-from-candidate flow, showing candidates for the user's choice. These tools are otherwise deprecated when create-design is listed; failure of create-design is not permission to fall back. generate-design does not support presentations despite its enum. The legacy outline-review widget and structured presentation generator are not exposed by this CLI. Do not substitute a non-branded generation when a brand-specific presentation workflow is unavailable. A selected brand template can still be copied with create-design-from-brand-template or autofilled.

Creation and layer separation return asynchronous jobs. When no Canva widget is shown (including CLI use), poll the corresponding get-create-design-async-job, get-generate-image-job, or get-separate-image-layers-job. Honor every returned wait interval and updated continuation token. Never start a replacement write because a job is pending. Stop on terminal failure; respect quota and moderation failures. Show completed results using the preview workflow below. For generated images, include the returned Canva upload link with the text Open generated image.

For get-create-design-async-job, save the non-empty job_id and continuation_token returned by create-design. Every poll must send that same job_id and the latest continuation_token; replace the token only when the previous poll returns a new one. A retry must resend both values unchanged. Never omit either argument, send it blank, or restart the polling chain with an older token.

Treat an explicit creation count as a hard mutation limit. After a create-design job succeeds, do not create, copy, or resize another design to correct its wording, layout, or quality unless the user explicitly authorizes another design. Report the mismatch and provide the Canva link so the user can correct the completed design in the Canva editor.

Upload and transform images

For attachments, local files, or generated files up to 256 MiB, use canva upload-file --file <path> --user-intent '<purpose>'. Its approval shows the selected file and an image preview for workspace images. The CLI obtains a single-use upload URL and sends the raw bytes after approval; use the returned resource IDs in later calls. Do not repeat an upload whose outcome is unknown. The low-level create-upload-url flow remains available for larger files: one raw-byte POST with Content-Type: application/octet-stream, no multipart, JSON, or base64. Never retry a consumed upload URL.

upload-asset-from-url and import-design-from-url accept already-public HTTPS sources only. Do not publish local or private files to use those tools. Uploading media does not place it into a design. create-design has no asset input parameter. When exact supplied media must appear and no listed tool can place it, explain that limitation and offer the Canva editor rather than using an unavailable workflow.

Use remove-background with an already-uploaded MEDIA reference to produce a new image with transparent alpha. It does not replace the scene or crop the subject. Use separate-image-layers with an uploaded image's asset_id to turn a flat graphic into a new editable design; the original stays unchanged. When the live catalogue has no tool capable of inspecting editable elements, say that layer verification is unavailable rather than calling an absent tool.

Read, edit, and organize

The current launch catalogue supports design and page discovery but not design content reads or editing transactions. Do not try remembered or future tool names for those workflows. Use get-design-pages when it is listed for saved page previews. When the user needs an unsupported content read or edit, provide the Canva link and explain that they must complete it in the Canva editor.

merge-designs combines or reorders whole pages. Obtain explicit approval of the exact operations before each call; deleting pages is permanent. copy-design and resize-design create new designs and preserve their source. Before autofill-design, inspect get-design-dataset or get-brand-template-dataset and match its field names/types. Set update_in_place only when the user requested overwriting that design.

Updating a brand template requires creating its draft, editing and saving that draft, then publishing it. The current launch catalogue cannot edit or save the draft, so provide the Canva link and ask the user to complete those steps in the Canva editor. Use publish-brand-template only after the user confirms the edited draft is ready and requests organization-wide publication. Publishing affects a reusable shared template.

Resolve Canva shortlinks before using designs. Confirm get-export-formats before export-design. Use help for current Canva product support questions, not to describe this CLI's capabilities.

Show results

Show a visual preview in chat alongside the returned Canva link when delivering designs or images. Save returned image content, or download a returned thumbnail URL with curl, to a file under workspace/. Inspect it with read, then attach it on its own line as ![Preview](sandbox://workspace/path/to/image.png). Preview each design in a small set; for a long deck, show representative pages and label their page numbers. Local attachments remain useful after signed preview URLs expire.

Check that the image shows the expected content; HTTP success alone does not rule out a blank or stale thumbnail. If needed, request a fresh thumbnail or export the saved design in a supported image format. Never commit unsaved edits just to obtain a preview, or present a saved export as an unsaved draft. If a usable preview is unavailable, explain that and keep the Canva link; do not repeat creation or editing to repair a preview.

When asked to list designs, omit --query and follow continuation tokens. This does not authorize background crawling or bulk indexing. Comments and replies are visible to collaborators; post only when the user requested them.