How headless site hosting works

Headless site hosting connects a person's request, an agent, a host interface, and a network that serves the result. Each part has a different job.

The parts and their roles

PartInput and operationOutput and limit
PersonThe person states what the site is for, approves payment, and decides what is public.The agent gets a goal and permission. The agent cannot complete a payment step for the person.
AgentThe agent writes or collects the files and calls the host.It produces a deploy request. It can only send what it has, and the files must be ready to serve.
Host interfaceThe API, MCP server, or CLI accepts files and lifecycle requests.It returns a URL, a site identifier, and an expiry, or an error. Size and rate limits apply.
Serving networkThe network stores the files and serves them over HTTPS.Visitors receive the files as sent. A static host runs no server-side code.
Domain and DNSThe host attaches a domain and issues a certificate.The site answers on that domain. A domain held elsewhere needs a DNS record that its owner adds.

An illustrative case

Consider a fictional bakery. Its owner wants a holiday menu page to share with customers.

This example describes a typical sequence. It is not a record of a real deploy.

  1. The owner asks the agent for the page. The agent writes an HTML file and a stylesheet.
  2. The agent requests an account from the host and receives a key and a payment link. The owner opens the link and completes the payment setup. This happens once.
  3. The agent sends both files in one request. The response contains a live URL and the time at which the site expires.
  4. The owner finds a wrong price. The agent sends the corrected file to the same site, and the URL does not change.
  5. The agent sets the site to expire on January 2. If the owner wants a permanent page instead, the agent pins it.
  6. If the owner wants the page on the bakery's own domain, the agent attaches the domain and tells the owner which DNS record to add.

The owner made three decisions: what the page says, whether to pay, and how long it stays up. The agent did the rest.

Where the work can stop

An unpaid account stops the first deploy. The agent must report the pending payment step instead of reporting a live site.

A request over the size, file, or rate limit fails. A useful error names the limit so that the agent can adjust the request.

A key with limited scope cannot buy a domain or change a plan. The agent needs a broader key from the person.

An expired site stops being served. An agent must not report a site as permanent unless it has pinned the site or confirmed the expiry.

A domain does not work until its DNS record exists and the certificate is issued. The agent can check the status and report it.

A site that needs a database or server-side code needs a different kind of host.