AI Agent Artifact Sharing in 2026: Claude, Cursor, Codex - Every Option Compared
By Marcin, founder of Markloop · Last verified July 2026 · Yes, Markloop is ours - the comparison stays honest anyway, and we say when a competitor is the better pick.
TL;DR: if your audience is inside your claude.ai org, native sharing is fine. If you need someone outside to just see it, a static host or an artifact-hosting tool (Send, ShareDuo, Anchorify) is the fastest path - pick Send if you want viewer analytics. If you need them to respond - comment, correct, approve, answer questions - none of those have a feedback mechanism; that's the job Markloop was built for, including pulling the feedback back into Claude Code over MCP. Cursor and Codex users hit the same wall from different sides: Cursor's shared canvases open externally but are read-only, and Codex Sites never leave the workspace.
The full comparison
| claude.ai publish (chat artifacts) | Claude Code share | Cursor shared canvas | Codex Sites | Static host (Pages/Netlify) | Send | ShareDuo / Anchorify | Markloop | |
|---|---|---|---|---|---|---|---|---|
| Viewable outside your org | Free/Pro/Max: ✓ Team/Ent: ✗ | ✗ never | ✓ (link) | ✗ workspace-only (preview) | ✓ (public) | ✓ | ✓ | ✓ |
| Viewer needs an account | no (public link) | yes - same-org claude.ai login | no | yes - workspace member | no | no | no | no - a link can allow comments (owner's toggle) |
| Anchored comments on the doc | — | — | — („not here yet") | in-workspace annotations | — | — | — | ✓ |
| Embedded questions → answers | — | — | — | — | — | — | — | ✓ |
| Feedback back into your agent | — | — | Design Mode - author only, in the IDE | annotations → Codex, in-workspace | — | — | — | ✓ MCP (Claude Code, Codex) or file export |
| Versioning with history | republish | ✓ versions (publisher picks which version viewers see) | republish snapshot | ✓ living pages | redeploy | ✓ link stays current | varies | ✓ + "addressed" tracking |
| Viewer analytics (who/when) | — | — | — | — | DIY | ✓ | varies | — |
| Custom domain | — | — | — | — | ✓ | ✓ | varies | — (roadmap) |
| Source file exposed to viewer | view-source | view-source | view-source | n/a | view-source | varies | varies | no - rendered view only |
| Price (creator) | plan incl. | plan incl. | paid Cursor plans | Business/Ent preview | free | paid, varies | paid, varies | $19/mo Solo, 14-day free trial (no card); reviewers free |
Facts about Anthropic features from the official docs; Cursor from the Shared Canvases changelog and docs; Codex Sites from OpenAI's June 2026 announcement (preview); Send/ShareDuo/Anchorify from their public sites, July 2026 - check current pricing there.
When each option is the right call
Native sharing - right when everyone's in the org
Zero setup, live updates, audit-logged. Its wall is absolute, though: no external viewer, ever, and no comments even internally. Fine for dashboards your team watches; wrong for anything a client must approve.
Cursor shared canvases - right for one-way external viewing
Publish a canvas and anyone with the link can open it in the browser - no Cursor account. That beats Claude's org wall for distribution. The gap is the same, though: viewers are spectators ("that capability is not here yet"), and Design Mode annotations work only for you, inside the IDE. When the viewer needs to talk back, you're in review-tool territory.
Codex Sites - right for living hubs inside the company
Sites stay updated as Codex works and members can point annotations at them - genuinely useful for internal dashboards. But they're a Business/Enterprise preview and shareable only inside the workspace, so a client or external expert never sees one.
Static hosting - right for public, one-way pages
Free and durable. But the URL is public to anyone who finds it, updates mean redeploys, and feedback is "reply to my email".
Send / ShareDuo / Anchorify - right for tracked distribution
These tools exist because the org wall is real, and they solve it well for the sales motion: a clean external link, and (in Send's case, per their site as of July 2026) viewer analytics - who opened it, when, for how long - plus custom domains. If your question is "did the investor open my deck", pick Send. Their shared gap: the viewer is still a spectator. No anchored comments, no questions, nothing flows back to the agent that wrote the doc.
Markloop - right when the doc needs a response
Markloop treats the artifact as a document under review, not a page to distribute. Reviewers (free, no Claude account - invited by email, or anyone with the link once you toggle comments on) comment on the exact element and answer the questions you embedded; you or your agent mark comments addressed in the next version; Claude Code and Codex push and pull over MCP so nothing is ever copy-pasted. The honest gaps in the other direction: no viewer analytics and no custom domain (roadmap).
FAQ
Can any tool make a Claude Code artifact itself public?
No. All external tools (including Markloop) work with the artifact's underlying HTML file, not the claude.ai URL. The artifact page itself never leaves the org - the file does.
Can I use Send or ShareDuo for feedback?
You can receive feedback about the page through whatever channel you already use (email, Slack) - but not on the page. None of the artifact-hosting tools offer anchored comments as of July 2026.
Can people comment on a Cursor shared canvas?
No - shared canvases are read-only for viewers (as of July 2026), and Design Mode annotations work only for the author inside the IDE. To collect comments from the people you shared it with, put the HTML your agent wrote into a review tool instead. Markloop connects to Cursor over MCP, so the push and pull happen without leaving the editor.
Does Markloop track viewers like Send does?
No. Markloop doesn't do open-tracking or viewer analytics - it's a review tool, not a sales-engagement tool. If you need both, they combine fine: Send for the tracked link, Markloop for the review round.
Need answers on the doc, not just eyes on it?
Share the artifact's HTML in Markloop, collect anchored comments and answers, and pull them back into Claude Code.
Start free - 14-day trial, no card required →Reviewers are always free.
Related: How to share a Claude Code artifact outside your organization · What are Claude artifacts? · How to get comments on a Cursor canvas