AI Agent Artifact Sharing in 2026: Claude, Cursor, Codex - Every Option Compared

By Marcin, founder of Markloop · Last verified October 2026 · Yes, Markloop is ours, and we say when a competitor is the better pick.

Which one to pick

• Only people in your org need it: native Claude sharing. On Team and Enterprise, teammates can comment and Claude Code reads and answers the threads.
• A client just needs to look: a static host (GitHub Pages, Netlify) or ShareDuo, or a Claude public link where your plan and admin allow one.
• You want to see who opened it: Send. It shows who opened the link, when, and for how long.
• People outside need to comment and discuss, and you want the comments back in your AI: Markloop. If your reviewers don't mind signing in, Anchorify also has inline comments and a CLI your agent can use.
• The baseline most agencies, consultants and freelancers use today: a PDF by email, or a Google Doc with comments. Docs is still the right call when the document is mostly text and the client already works in Google Docs or wants to edit it directly. It gets awkward when your AI made a designed HTML page, or when you want the comments back in your AI without copying them over.

The comparison in three tables

Who can open it

Who can comment, and does it reach your AI

Versions, tracking and price

This is the table where Markloop loses most rows.

Facts about Anthropic features from the Claude Code docs and the Claude Help Center; Cursor from the Shared Canvases changelog and docs; Sites from OpenAI's June 2026 Codex announcement and the July 2026 ChatGPT Sites launch; Send, ShareDuo and Anchorify from their public sites, September 2026 (check current pricing there). *Per the Claude Code docs, a public Claude Code link opens with no claude.ai sign-in. The Help Center says everyone needs a Claude account to open a shared artifact, except a legacy artifact published from a chat; chat artifacts made since 16 September 2026 use the Share menu instead of Publish.

What changed since July 2026

• Claude Code artifacts can go public: a public link on Pro and Max, and on Team and Enterprise once an Owner turns on External sharing. A page that calls connectors can be shared too, but a viewer who isn't signed in, or who is outside your org, sees it without the live sections. In the Claude app, artifacts that use Claude or connect to your apps can't use "Anyone with the link".
• Comments on Claude artifacts: org members, and people invited by email (beta, Claude account required), can comment. Claude Code reads the threads and can reply. Public links take no comments.
• Cowork uses the same sharing: Cowork artifacts made since 19 August share like any other artifact; older live artifacts stay inside the org.
• Codex Sites became ChatGPT Sites (9 July): beta on paid ChatGPT plans, with public publishing and custom domains; email invites for outside people followed in September.
• A correction on Cursor: we said in July that anyone with the link can open a shared canvas. Cursor's docs say only members of your Cursor team can open one, so we fixed the table.

When each option is the right call

Native sharing - right when everyone's in the org

No setup, and the artifact updates in place. On Team and Enterprise, teammates can comment on a Claude Code artifact shared inside the org, and Claude can read and reply to those threads. For outsiders it's narrower: a public link works on Pro and Max, and on Team and Enterprise only after an Owner turns on External sharing. People who only have the public link can't see or add comments, and people you invite by email (beta) need a Claude account. Good for dashboards and reviews inside your team, awkward when a client has to respond. (Guide: how to share Claude artifacts outside your organization.)

Cursor shared canvases - right for one-way viewing inside your team

Publish a canvas and your teammates can open the link in the browser. Only members of your Cursor team can open a share, so it won't get a canvas to a client. Viewers can't comment, and Design Mode annotations work only for you, inside the IDE. (Guide: how to get comments on a Cursor canvas.)

ChatGPT / Codex Sites - right for living hubs, now public too

A Site stays updated as Codex works on it, which suits internal dashboards. ChatGPT Sites is in beta on paid ChatGPT plans (not Free or Go). A Site can go out on a public URL or a custom domain (on Enterprise an admin has to allow public publishing), and owners can invite specific people outside the workspace by email, who then sign in to open it. Annotations are for the author to direct Codex. We couldn't find a way for outside viewers to leave feedback on the page. (Full guide: ChatGPT Sites explained.)

Static hosting - right for public, one-way pages

Free, and the page stays up as long as you like. Anyone who finds the URL can open it, each update is a redeploy, and feedback arrives by email.

Send / ShareDuo / Anchorify - right for tracked distribution

You get a clean external link, and with Send (per its site) viewer analytics: who opened it, when, for how long, plus a custom domain. If your question is "did the investor open my deck", pick Send. With Send and ShareDuo the viewer can only look: there are no comments on the page, and nothing goes back to the agent that wrote the doc. Anchorify has moved toward review: signed-in viewers can comment inline and suggest edits, and your agent can list and reply to comments through its CLI or REST API.

PDF by email or Google Docs - what most client work runs on today

Everyone can open a PDF, and nobody has to learn a new tool. Feedback comes back as a reply email or a marked-up PDF, and each new version is another attachment. Google Docs keeps comments in the margin next to the text, with version history, and clients can suggest edits themselves. If your AI wrote plain text, a Doc is often the simplest choice. A designed HTML page is harder: converting it to Docs loses the layout (we tested what survives). With either one, getting the comments back to your AI means copying them over.

Markloop - right when the doc needs a response

Take the artifact's HTML file (download it from Claude, or use the file your agent wrote in Claude Code, Cursor or Codex) and ask your AI: "Publish this file to Markloop and give me a share link with commenting on." People comment on the spot they mean and reply to each other in threads. When you're ready, your AI pulls the comments, makes the changes and publishes the next version at the same link, marking which comments it addressed.

Where it falls short: no viewer analytics, no link passwords (links can expire and be revoked), no custom domain yet, and it's built for documents, not live multi-page sites or apps. It's paid for you as the creator, while reviewers never pay. The homepage FAQ covers who can see your link and where your data lives, and how to write a client proposal with AI and get comments on one link walks through a full round with a client.

No setup needed to try it: download the HTML file your AI made and upload it to Markloop. Or connect Claude, ChatGPT, Cursor or Grok once (setup is a one-time step, shown in the app after signup) and ask your AI to publish. Watch the 2-minute walkthrough.

FAQ

Related: How to share Claude artifacts outside your organization · What are Claude artifacts? · Claude live artifacts are legacy · How to get comments on a Cursor canvas