Alternative · Claude Code Artifacts
A Claude Code Artifacts alternative for what happens after you share it
Artifacts can't be shared outside your Anthropic org.
That aged like milk. Anthropic shipped public sharing links in the week of 13 July 2026, so we've struck it rather than hidden it.
Honestly? We assumed they'd get there — and they'll keep getting there. "Can you share it" was never the interesting question. "What happens after you share it" is: who opened it, when it stops working, whether you can lock it to one client instead of everyone, and whether anything checked page 3 for a live API key.
So "can you share it" is settled — anyone with the link can open it, no sign-in. What hasn't changed is everything after that: there is no viewer analytics (the org audit log records who *published*, not who read), no per-link expiry, no way to gate a link to one client's email address, and it is single-page HTML or Markdown only, capped at 16 MiB — no PDFs, no images, no multi-file bundles. And nothing checks the output for credentials on the way out. freo.cloud is the delivery layer instead: it takes what your AI agent makes — reports, dashboards, HTML pages, PDFs, images, whole static sites — and gives it a link exactly the people you choose can open, with no account required. You pick the access level per document (public, an unguessable link, email-verified, or a named allowlist) and change it later without changing the link. You can see which person opened it and when. Links retire themselves on a schedule. Every publish is scanned for leaked credentials and blocked before the link exists. And freo isn't tied to one vendor — it publishes from ten coding agents (Claude Code, Codex, Cursor, Windsurf, opencode, Antigravity and more) and from chat apps like Claude.ai and ChatGPT over MCP.
“…On Pro and Max plans, a public link is the only way to share an artifact.”
freo.cloud vs Claude Code Artifacts, side by side
| freo.cloud | Claude Code Artifacts | |
|---|---|---|
| Gate a link to one outside person | ✓Yes — email-verified, or a named allowlist | No — tiers are private · named people in your org · whole org · fully public. No external-email tier |
| Scans for leaked credentials | ✓Yes — the publish is blocked before the link exists | No — not documented |
| Viewer needs an account | ✓No — at most a one-time email code | No for public links; org sharing needs a claude.ai sign-in |
| Per-document access control | ✓4 tiers: public · unique link · email-verified · named allowlist | Private · named org members · whole org · public |
| View tracking | ✓Per viewer — who opened it and when | None for viewers (the audit log records publish/share/delete) |
| Auto-expiry / dormancy | ✓Yes — hard TTL + inactivity auto-private | No per-link expiry (org-wide retention policy only) |
| Works across agents | ✓10 coding agents (Claude Code, Codex, Cursor, Windsurf…) + Claude.ai & ChatGPT via MCP | Claude Code only |
| Content types | Markdown, HTML, PDF, images, multi-file site bundles | Single-page HTML / Markdown, ≤16 MiB |
| Plan required | ✓Any freo account — free during early access | Pro · Max · Team · Enterprise (off by default on Team/Enterprise) |
| Vendor lock-in | ✓Cross-vendor; export any time | Anthropic-hosted, no custom domain (HTML is exportable) |
Claude Code Artifacts details reflect Anthropic's published documentation as of July 2026; “not documented” means Anthropic publishes no detail, not that the feature is confirmed absent.
Why freo is the alternative
Nothing leaks by accident
Standard publishes are scanned for API keys and credentials and blocked before the link exists — text, PDF text and image metadata, on every publish path. Anthropic does not check at all.
Locked to one person, not one org
Email-verified or a named allowlist reaches a client who isn't in anybody's directory — and you can change it later without changing the link.
Any agent, not one vendor
Ten coding agents — Claude Code, Codex, Cursor, Windsurf, opencode and more — plus Claude.ai & ChatGPT over MCP. Same link, same controls.
When Claude Code Artifacts is the better choice
We're not pretending Claude Code Artifacts is bad — for the right job it's the path of least resistance. Reach for it when:
- →The audience is your own teammates, for quick internal review — or the whole internet, and you don't care who reads it.
- →You're already on a Claude plan and want zero setup.
- →You want the artifact built live from full session context — diffs, tool output — and updated in place.
- →You want viewers' own MCP connectors to power the page when they open it. That's genuinely clever, and freo has no equivalent.
- →Keeping the work inside Anthropic's security boundary is the point (no new vendor for IT to vet).
And they complement freo: build the work in Claude Code Artifacts, then export it and publish through freo when you need an external, access-controlled, tracked link.
Questions
Can I make a Claude Code Artifacts page public?
Yes — Anthropic added public sharing links in the week of 13 July 2026, and anyone with the link can open it without signing in. This page used to say otherwise; we've struck that above rather than deleted it. What you still can't do is point that link at one person outside the org, see who opened it, or have it expire — which is where freo comes in.
Does Claude Code Artifacts check for leaked API keys before publishing?
No. OpenAI's docs advise you to “keep secret values out” of Site content, which is guidance rather than a check, and Anthropic documents nothing here either. freo scans every standard publish — text files, PDF text, and image EXIF metadata, including inside multi-file bundles — and blocks it by default on every publish path, showing you exactly what was found so you can fix it. (Opt-in end-to-end-encrypted documents are the one exception, and for an honest reason: we receive ciphertext we cannot read.) To be straight about the wider market: we are not the only tool that scans. Stacktree does too, though it blocks by default only on its MCP path. And Netlify and GitHub both scan — but their gate is a build or a git push, so uploading an already-finished PDF or prebuilt folder, which is exactly what an agent hands you, sails straight past it. Agents paste real credentials into example blocks and logs constantly; that is the failure mode we built for.
Does the person I share with need an account?
With freo, no. Public and unique-link documents open with no sign-in at all; email-protected and allowlist documents use a one-time email code — never a password or a new account.
Do I have to switch agents to use freo?
No. freo is cross-agent — paste one setup prompt into any of the ten supported agents (Claude Code, Codex, Cursor, Windsurf, opencode and more) and it publishes from whichever you already use. And you don't need a terminal at all: Claude.ai and ChatGPT connect to the same account over the MCP server with one-click OAuth.
Comparing the other one too? freo.cloud vs ChatGPT Sites → · See every supported agent →
Publish your agent's work where anyone can see it
Connect your coding agent in one prompt — or your chat app in one click — and share a link you control: public, private, or locked to named people. Free during early access.
Get your space →