C CastHTML
Share AI artifacts

Share Claude artifacts & AI HTML, properly

Artifacts look great inside the chat — and fall apart when you paste them anywhere else. CastHTML publishes them as sandboxed links, and its MCP server lets your agent publish them itself.

Publish straight from your agent — the MCP server

CastHTML exposes a native remote MCP server at https://casthtml.com/mcp with five tools: publish_report, update_report, list_reports, get_report, and delete_report. Create an API key on your history page, then add either config — or see the dedicated MCP server guide for every client's setup, tool annotations, and the verified RemoteMCP.net registry entry:

# Claude Code — one command
claude mcp add --transport http casthtml https://casthtml.com/mcp \
  --header "Authorization: Bearer cast_live_YOUR_KEY"
# Any MCP client (JSON config)
{
  "mcpServers": {
    "casthtml": {
      "type": "http",
      "url": "https://casthtml.com/mcp",
      "headers": { "Authorization": "Bearer cast_live_YOUR_KEY" }
    }
  }
}

Works with whatever produced the artifact

Claude artifacts & slides

Claude Code artifacts, reading-slide artifacts, and dashboards paste straight in — layout intact inside the sandbox.

Gemini-hosted HTML

Asked Gemini to build you a page? Host it properly instead of screenshots — paste the HTML, get the link.

ChatGPT canvases & code

Any generated HTML or Markdown works the same way — the renderer doesn't care which model wrote it.

Iterating with the agent

update_report edits in place, so "publish it, then tweak the colors" is a single conversation.

Frequently asked questions

How do I share a Claude artifact with someone?

Copy the artifact HTML into CastHTML and publish — the link preserves the rendered page inside a safe sandbox. Or let Claude publish it directly via the CastHTML MCP server.

Can an AI agent publish reports itself?

Yes. CastHTML runs a native remote MCP server at https://casthtml.com/mcp. Add it to Claude (or any MCP client) with your API key and the agent can publish, update, list, and delete reports.

Does it work with Gemini or ChatGPT output?

Any HTML page, Markdown document, or ZIP-packaged static site an AI produces can be published — the workflow is the same regardless of which model generated it.

Are shared AI artifacts safe to open?

Yes — sandboxed rendering is the default. Scripts and network calls in the artifact are blocked unless a signed-in publisher deliberately enables trusted mode.