Deploy the website your AI built. Review the release first.
DeployBeacon gives AI-built static websites a publishing workflow: built files, inspection evidence, a private preview and an explicit human decision before a compatible MCP assistant publishes.
Start with the files you already built
Ask your coding assistant to produce a production build and a ZIP of the output. Upload it in your invited workspace. This route works without granting an assistant access to your account or asking it to handle credentials. The public ZIP handoff instructions describe the boundary: prepare files with permission, return the ZIP and let you decide what to publish.
Connect a compatible MCP client
Connected clients use the same projects, permissions, inspections and release history as the dashboard. A compatible assistant can read evidence and submit built files, propose publication or recovery, then execute the proposal after you approve it. It cannot approve its own proposal.
Connect with OAuth consent or the documented token route. Select the workspace and only the permissions needed. Client-specific compatibility is still being validated; this beta does not claim universal or certified support for every AI chat application. Read the API, CLI and MCP connection guide.
What do you approve?
The exact project, action and inspected or retained release. Review the proposal in Agent approvals within 15 minutes. Once approved, the requesting connection has a further 15 minutes to execute that action. Check the recorded outcome before assuming the website changed.
Legacy API and CLI tokens can have direct deployment-write permissions. Grant those only to tools you trust; a token is not interchangeable with an approval-gated MCP proposal.
Small built websites can be submitted directly
The MCP files_inspect tool accepts up to 50 files and 512 KiB of decoded content, within the 1 MiB request limit. It creates a private inspection rather than a live website. Larger outputs use the ZIP route. Credential and archive checks apply before publication.
Keep a recovery path
A failed candidate does not replace the selected live release. Retained successful releases can be restored. Private previews let you check a proposal before going live, and publication remains subject to project permissions and workspace allowances.
Read about credentials and access and preview and restore a release.
Explore the product and guides
Access is invitation-only. No payments are collected during the private beta. Sign in to your invited workspace.