Technical note

AI-native authoring: when the editor understands its own language

Why AI in Litepress is not a chat beside the editor, but an MCP-assisted workflow where the agent reads current Markdown instructions based on enabled extensions and integrations.

Published

May 3, 2026

By Simon Zajdela

AIAuthoringLitepressPublishing
A structured editor, system rule registry, and AI context layer connected in a publishing workflow.

AI needs the current rules of the system

Litepress uses an MCP workflow to make the publishing language visible to the agent. Before the agent writes or changes Markdown, it can read the current Markdown instructions exposed by the system.

Those rules are not just static documentation. The available instructions can depend on the current site configuration and enabled integrations. If newsletter integration is not enabled, the agent does not need to be bothered with newsletter block syntax. Less context noise, fewer tempting wrong moves.

The model must not guess syntax

If a system has custom Markdown extensions, AI must not guess what they look like. Guessing is fast, confident, and often almost correct. Almost correct is a dangerous category in structured content.

An AI agent needs to know which extensions exist, which parameters they accept, which fields are required, which values are allowed, and what valid output looks like. That knowledge has to come from the system.

Related thinking