Keep the AI workflow connected to the finished file.

Generate a deliverable you can inspect

Ask your AI tool for a static HTML deliverable and say whether resources should be embedded or packaged separately. Define the expected screen sizes and interactions. For a product proposal, require a readable structure; for a report, identify the data source and calculation assumptions rather than relying on attractive charts alone.

Keep credentials and personal data out of the generated file. Check buttons, links, charts and mobile layout yourself. A convincing interface may still contain placeholder data or controls without a real backend. Make those boundaries clear in the deliverable before anyone reviews it.

Upload, then check the hosted result

Use the workspace to upload a single HTML file or a complete ZIP. Check the hosted page for missing assets and blocked external dependencies. The initial project remains private. Create an access grant only after confirming that the intended files, interactions and information are present.

Use one project for successive revisions of the same deliverable. Update its content instead of creating a new project for every change. The address and existing grants remain intact. Keep a separate local source history if you need to recover older versions; hosted replacement is not a version-control system.

Use MCP to connect a compatible AI client

Model Context Protocol, or MCP, lets compatible AI clients call defined tools. In Faway, those tools can inspect your space, list projects, upload files and update existing content. MCP transports the operation; it does not generate the HTML or supply a missing backend.

Create a separate MCP credential in your account and configure the local adapter with explicitly allowed directories. The adapter can upload HTML or ZIP files up to 50 MB. Remote MCP accepts HTML strings up to 1 MB and cannot read a path on your computer. Grant-management tools need an additional permission; a Google login token or visitor code is not an MCP credential.

For a retry, reuse the same idempotency key so a lost response does not become a duplicate project. For an update, provide the project ID and current revision. If another update changed the revision, inspect the current project before replacing it again.

Apply the workflow to distinct deliverables

For an interactive prototype, keep the project address stable through review rounds. Ask the AI to revise a specified interaction, inspect the local result, replace the hosted project and have the reviewer refresh the same page. Use separate projects when reviewers need different versions.

For a data report, check the source, totals and labels before upload, then share only the reviewed report through a time-limited grant. For a product proposal, package screenshots and diagrams alongside the HTML so the document survives delivery without references to your local files.

These are individual creator workflows, with no team collaboration or payment step required. Create an account with verified email and a password, or continue with Google. You can link an additional sign-in method from Space settings. Google users with a linked email can set an email sign-in password through password reset.