> ## Documentation Index > Fetch the complete documentation index at: https://00doc-web.vercel.app/docs/llms.txt > Use this file to discover all available pages before exploring further. # Publishing a document > Upload HTML from the dashboard or publish via MCP from Claude and Cursor. # Publishing a document You can publish to Unfurl in three ways. Pick the one that matches your workflow. Go to the [dashboard](https://unfurl.pub/dashboard), create a new document or open an existing one. Drag an `.html` file onto the upload zone, or use the file picker. The preview updates and the share link reflects the new version. Connect [MCP](/core-concepts/mcp-overview) with endpoint `https://mcp.unfurl.pub/mcp`, then ask in natural language: > "Publish this as an HTML report titled 'Weekly KPIs' to Unfurl" Claude calls `publish_document` with `title`, `html`, and optional `note`. You get back `document_id`, `version_id`, and `share_url`. Add the same MCP URL in Cursor settings, then prompt from the AI chat: > "Generate a one-page HTML spec for the billing API and publish it to Unfurl" Cursor invokes `publish_document` the same way Claude does. ## HTML quality * Prefer **valid, complete documents** (``, semantic structure). The sanitizer strips risky markup; broken or empty pages are hard for readers and agents to use. * **Inline CSS** or linked stylesheets you control work best for predictable rendering in the preview and on `/d/[token]`. * **Test locally** in a browser before upload if you use heavy scripts — the public view runs in a sandboxed iframe with a tightened CSP. **Titles matter.** Use a specific title ("Q3 renewal — Contoso") instead of "Document 2". It shows in the dashboard, in history, and when agents fetch metadata. ## Prompts for AI-generated HTML Give the model: * **Audience** (executive, engineer, customer) * **Structure** (sections, tables, callouts) * **Output format:** "Return a single HTML file with embedded CSS" Then ask it to publish via Unfurl so you do not paste multi-kilobyte HTML manually.