How it works

1. Build for people and agents

Sites start from structured content. That content is rendered into fast, semantic, accessible HTML for people, and into Markdown, JSON, and structured data for agents.

• Semantic HTML5 landmarks and a logical heading hierarchy
• JSON-LD structured data (Schema.org) on every page
• Markdown versions of every page, plus llms.txt and llms-full.txt
• A sitemap, a JSON content index, and server-side search
• Responsive layouts with light and dark themes

2. MCP-native by default

Each site can expose a Model Context Protocol server. Agents connect over Streamable HTTP and use tools to work with the site directly: listing pages, reading content, and searching.

Read-only tools are open to anyone. Tools that change content are designed to create proposals for review, not to publish directly.

3. Human-in-the-loop change management

Agents are useful editors. They can fix typos, update outdated details, and draft new pages. But publishing is a decision for people. In sight4agents, every agent edit follows the same path:

1. Propose

   An agent submits a change proposal with the edited content and a short explanation of why.

2. Preview

   The proposal is rendered as a diff and a preview, so reviewers see exactly what will change.

3. Review

   A person approves it, asks for revisions, or rejects it. Nothing is live yet.

4. Publish

   Only approved changes are published.

5. Record

   Every step is logged so changes can be traced and rolled back.

Who does what