HTML5Deckary Canvas

Agents build HTML decks you can't edit, now you can

Claude Code builds the deck. You open the visual editor, tweak what you want, annotate the rest, and hand it back. The loop keeps going without starting over.

No account required to get started

Our users work at companies like

Deckary helps consulting & industry firms to build better presentations

AirbusWarner Bros. DiscoveryUFCApolloS&P GlobalEYGELongchamp
AirbusWarner Bros. DiscoveryUFCApolloS&P GlobalEYGELongchamp
AirbusWarner Bros. DiscoveryUFCApolloS&P GlobalEYGELongchamp
AirbusWarner Bros. DiscoveryUFCApolloS&P GlobalEYGELongchamp
Deckary Canvas

Your agent creates HTML, Canvas makes it editable

The HTML deck your coding agent writes, opened in Deckary Canvas with an interactive editing layer on top.

HTML slide generated by a coding agent
Agent outputPlain HTML, exactly as generated
The same deck open in the Deckary Canvas editor
In Deckary CanvasSame deck, now editable

How the round-trip works

Other tools stop at generation. Deckary Canvas is built around the back-and-forth that actually finishes a deck.

1. Your agent builds the deck

Paste one prompt into Claude Code, Cursor, or Codex. The agent initializes the project, writes the slides as plain HTML, and starts a local preview.

2. You edit what you can see

Open the visual editor. Click text to rewrite it inline, drag elements to reposition, resize boxes — the things that take three prompts to explain take three seconds to do.

3. You annotate the rest

Pin comments to the exact element they're about. Annotations are saved into the project itself, not lost in a chat scrollback.

4. The agent picks it back up

Your agent reads the annotations from the local project, applies precise operations through MCP tools, and marks them resolved. The loop repeats until the deck is done.

The whole deck is a folder

my-deck/

deck.deckary.htmlthe slides, plain HTML
theme.cssdesign tokens and layout
assets/images and media
deckary.canvas.jsondeck manifest
.deckary/annotations.jsonyour feedback, agent-readable

No database, no proprietary binary. Your agent reads and edits these files directly; the visual editor writes back to the same source.

The part HTML decks were always missing

Claude Artifacts are throw-away. Raw HTML exports are frozen. Deckary Canvas adds the edit layer — for you, for your agent, and for the round-trip between the two.

Actually editable

Open the visual editor after Claude builds the deck and tweak text, layout, and content without touching code. No export step, no format conversion.

The loop keeps working

Edit visually, then hand back to your agent. Stable element IDs and structured operations mean Claude can continue from where you left off — not rewrite from scratch.

Annotations as project state

Leave feedback in the visual editor. Your agent reads it from the local project, acts on it, and marks it resolved — without losing it in chat history.

Plain files throughout

The deck is deck.deckary.html, theme.css, and a few sidecars. Your agent reads and edits them directly. No hidden state, no proprietary format.

MCP tools for precise edits

A project-scoped MCP server lets the agent inspect slides, apply targeted operations, validate, and export — instead of regex-rewriting raw HTML and hoping.

Present, export, share

Present straight from the browser, export static HTML, PDF, or PNG slides, or publish a hosted view-only link when the deck is ready to send.

Where the other approaches stop

Every way of getting a deck out of a coding agent breaks down at the same point: the first edit after generation.

After generationDeckary CanvasClaude ArtifactsRaw HTML filesPowerPoint export
Edit visually without re-prompting
Source lives on your machine
Agent continues without rewriting from scratch
Feedback the agent can read and resolve
Stable element IDs for precise edits
Browser-native layout and motion
Works with any coding agent

The deck your agent already builds, plus the edit layer it was missing.

Deckary Canvas

Deckary Canvas

Agents create HTML you can't edit, now you can

Open Claude Code, Codex, Cursor, or any coding agent. Paste one prompt and let it install Deckary Canvas, initialize the local deck, and start the preview — then open the visual editor and take it from there.

Open a project folder

Start where the deck should live, next to the files your agent can edit.

Drop in the install prompt

Ask your coding agent to run the Canvas package, initialize the deck source, and start preview.

Review in the browser

Iterate with comments, validation, and exports while the source stays local.

Install prompt
Paste into your coding agent
Create a Deckary Canvas presentation in this folder.

1. Init and launch (the sample is live immediately — don't wait on content):
   - Claude Code:   npx @deckary/canvas init . --agent claude-code --start --detach
   - Claude Cowork: npx @deckary/canvas init . --agent cowork --start --detach
   - Other agents:  npx @deckary/canvas init . --agent generic --start --detach
   `--detach` backgrounds the preview so your turn isn't held. Then run `deckary-canvas validate` and give me the preview URL.

2. Ask one question: should you (a) match a deck I'll drop here (.pptx/.pdf/.key → theme.css), (b) build fresh on a topic I name, or (c) keep the sample?

Stay local-first: edit deck.deckary.html, theme.css, deckary.canvas.json, assets/, and .deckary/annotations.json directly. Don't move the deck to another app, require login, or use screenshots instead of source edits.

Use the Deckary Canvas MCP tools (validate, preview/export, annotations, CanvasOperation) when available — a fresh .mcp.json works only after approval (`/mcp` or restart); until then use the CLI (validate / preview / export), which needs none.

The agent handles setup, preview, validation, and export.

Local-first by design

The deck is a folder of plain files on your machine. Your agent edits the source directly, validation keeps the round-trip reliable, and nothing has to leave your laptop until you decide to share.

No account is needed to initialize, build, and preview. A Deckary account unlocks the visual editor and hosted export — the parts that involve us, not your files.

Git-friendly

Plain text files diff, branch, and version like any code. Review deck changes in a pull request if you want to.

No lock-in

The deck stays standard HTML and CSS. It opens in any browser, with or without Deckary.

Runs locally

Init, build, preview, validation, and export all run on your machine. No upload required to work on a deck.

Build it with AI. Edit it like a human.

Paste the install prompt into Claude Code, Cursor, or any coding agent. Then open the visual editor and take it from there.

Frequently Asked Questions

Everything you need to know about Deckary Canvas