← Back to Blog

How to Use Claude Artifacts in ChatGPT, Cursor, and Gemini

How to Use Claude Artifacts in ChatGPT, Cursor, and Gemini

Most companies don't run one AI anymore. Engineers work in Claude Code or Cursor, the sales team got ChatGPT because that's what IT bought first, and the data people live in Gemini. The spend numbers say the same thing: Menlo Ventures' 2025 enterprise survey put Anthropic at 40% of enterprise LLM API spend, OpenAI at 27%, and Google at 21%. Nobody owns the whole room.

That's fine until someone builds something good. A product lead asks Claude for a launch plan or a pricing calculator, it comes out as an artifact, and a colleague who works in ChatGPT wants to change it. They can't open it. They can get a copy, and the copy starts drifting from the original the day it's made.

Here's what works today, what each workaround costs you, and what we'd do instead.

Can ChatGPT Open a Claude Artifact?

No. Claude artifacts live on claude.ai. Anthropic's artifacts help page says everything you make is saved to the Artifacts tab in the Claude sidebar and starts private. There's no import path into ChatGPT, and Anthropic offers no self-hosted version, so you can't put the artifact somewhere ChatGPT could reach on its own.

What you can do is share a link. Per Anthropic's sharing docs, an "Anyone with the link" share opens for people outside your org, and most open without a Claude account. So your ChatGPT colleague can look at it in a browser. Looking is all they get, though. For viewers outside the org or not signed in, the parts that use connected apps or Claude don't work. Docs are stricter still: they open only for people signed in to Claude, and those people can only view them. On Team and Enterprise plans artifacts stay inside the org until an Owner turns on External sharing.

It runs the other way too. Work in ChatGPT Canvas stays in ChatGPT, and Claude can't open it. Neither app imports from the other.

How to Export Claude Artifacts (and What Each Export Loses)

If you want the artifact somewhere else, export is the supported route. The export options depend on the template:

  • Docs: Word, PDF, Markdown, or Google Docs.
  • Slides: PowerPoint, PDF, or Google Slides.
  • Designs: .zip, PDF, PowerPoint, or standalone HTML.
  • Motion: MP4.
  • Sheets: .xlsx or CSV, according to Anthropic's Sheets and Data launch post.
  • Legacy artifacts (made in chat before September 16, 2026): view the code, copy it, or download it.

Every one of those gives you a file. A file is a snapshot, and the things that made the artifact more than a file stay behind in Claude.

Stored data is the first to go. New artifacts can store up to 100 MB and 25,000 records by default, and an export captures what the artifact shows, not the records it keeps. Anything that calls Claude stops working, since that feature runs on each viewer's own Claude plan and sign-in. Connected apps go too: an artifact that pulls from a connected app uses the viewer's connections, and a PDF has no viewer. The share link, and the list of people you gave it to, stay on claude.ai.

One trap for legacy artifacts: unpublishing one permanently deletes its stored data. If you're cleaning up old artifacts, export first.

Editing a Claude Artifact With Another AI

The usual way to edit a Claude artifact with another AI goes like this. You export it or copy the code. You paste it into ChatGPT or open it in Cursor. You make your changes. Then you either upload the result back to Claude by hand, or (far more often) you don't.

Now there are two versions. Someone edits the Claude one on Tuesday, someone else edits the ChatGPT one on Wednesday, and by Friday nobody can say which is current. Spreadsheets have been doing this to finance teams for decades, and AI makes it faster because each copy is so cheap to make.

If you want to move Claude artifacts to another platform for good, the export route is fine. Export once, call the copy the new original, and stop editing the old one. It breaks down when the file needs to keep living in more than one AI at the same time, which is the normal case in a company that runs several.

ChatGPT Canvas and Claude Artifacts in One Place

Each chat app has its own version of this feature. ChatGPT Canvas is a side-panel editor for writing and code, and the work stays in ChatGPT. Gemini Canvas works the same way and hands work off to Google Docs, Slides, or Colab. Open WebUI and LibreChat get closer, since they're self-hosted chat apps that work with Claude, GPT, and local models, and they can render HTML artifacts in a chat. But the artifact belongs to that chat. There's no shared company folder.

So no chat app is going to be the one place. We compared the options side by side in our post on Claude artifacts alternatives, and the pattern holds across all of them: the file lives wherever it was made.

Our fix is to move the file out of every chat app. Put shared files in a store that sits outside Claude, ChatGPT, and Gemini, and let each AI read and write it over MCP. MCP, the Model Context Protocol, is the open protocol Anthropic introduced in November 2024 for connecting AI apps to outside tools and data, and Claude, ChatGPT, Cursor, Gemini CLI, and Codex all support it as clients. It's the same argument we made about models in Why Enterprises Need LLM-Agnostic Architecture: the thing you care about keeping shouldn't live inside the vendor you might swap.

Sharing AI Artifacts Across Claude and ChatGPT Over MCP

Anchoris one such store. It describes itself as a file system for AI artifacts: HTML pages, Markdown docs, presentations, apps, and live tables, kept in folders, each with its own link. Agents read and write over MCP, and people can edit in the browser. Per Anchor's docs, pages are stored as HTML, documents as Markdown, and tables in PostgreSQL, so the files are in formats any model already reads. Anchor also says it keeps a version history per edit, records which agent or person made each change, and lets you restore an earlier version.

The setup differs a little per agent. All of the steps below come from Anchor's own documentation.

Claude

On claude.ai, go to Customize, then Connectors, then Browse connectors, and add Anchor. After that Claude can create and edit files in Anchor from any chat.

ChatGPT

Anchor's ChatGPT setup page has you install Anchor from ChatGPT's plugin directory and sign in. Then you type @Anchor in a chat to bring it in.

Cursor and Gemini CLI

For coding agents it's a remote MCP server. Anchor's Cursor setup guide and its Gemini CLI page follow the same steps:

  1. Add an MCP server with the URL https://mcp.anchor.cc/mcp.
  2. Sign in when the agent prompts you.

Codex and OpenCode take the same URL. Anchor lists any MCP-compatible agent as supported, so the same steps apply to a client it doesn't name.

An Example: One File, Three AIs

Here's how it plays out, as an illustration rather than a client story. A product lead asks Claude to draft a launch plan and save it to Anchor as a Markdown doc. The ops lead, who works in ChatGPT, types @Anchor, opens the same doc, and adds rollout dates. An engineer in Cursor reads it while building the feature flag and updates the status table at the bottom. Nobody exported anything, there's one file, and the history shows which AI made each edit.

Anchor has liabilities, and you should weigh them. It isn't open source. It's a young product. It has nothing like Claude Data's certified metric definitions. And artifacts that call Claude from inside the page don't carry over, so you rebuild those features as apps. We're also related to Anchor (see the disclosure below), so read our view with that in mind.

When to Leave an Artifact in Claude

Plenty of artifacts should stay where they are. If everyone who touches the file uses Claude, there's nothing to fix. If the artifact calls Claude for each viewer, that feature only exists on claude.ai, and rebuilding it is work you should only take on with a reason.

Claude Data is the clearest case. Anthropic's launch post describes certified metric definitions and live sharing, where viewers see approved query results without needing access to the connector. If your dashboard depends on that, keep it in Claude. A deck you'll export to PowerPoint once and present also belongs there. A one-way handoff is what export is for.

The problem is narrower than "artifacts." It's the files that several people keep editing from different AIs. Those are the ones we'd move out of the chat apps, before the second copy shows up.

Disclosure: Neural Bridge and Anchor are related companies.