AI Authoring Workflow

In Anydocs, the default AI workflow is: install and register the MCP server, read project constraints, then write through MCP with explicit page, navigation, and status boundaries.

Steps

Install and register the MCP server

Before letting an AI assistant participate in authoring, register the Anydocs MCP server in your runtime environment. Refer to the Installation page for quick-install commands. Common registration configs:

codex mcp add anydocs -- npx -y @anydocs/mcp

claude mcp add -s user anydocs -- npx -y @anydocs/mcp

Initialize the project and prepare the agent entry file

Create the project with `init`; if you want a minimal guide file for a specific agent, include `--agent` during initialization.

Read the project contract first

Have the agent call `project_open` before writing so it can inspect allowed blocks, templates, languages, resources, and authoring guidance.

project_open({ projectRoot: "/path/to/project" })

Establish page and navigation context

Continue with `page_list`, `page_get`, `page_find`, and `nav_get` so the agent understands the current content and navigation structure before deciding what to change.

Write through page and navigation tools

Create or update content through template tools, page tools, and navigation tools instead of editing `pages/<lang>/*.json` or `navigation/*.json` directly.

Keep publication as a separate action

After the content is reviewed, call `page_set_status` to move the page to `published`; do not change status through `page_update`.

page_set_status({ projectRoot: "/path/to/project", lang: "en", pageId: "guide", status: "published" })

Verify both human and AI delivery outputs

Finish with `preview` or `build` and confirm the reader pages, search indexes, `llms.txt`, `llms-full.txt`, and `mcp/*.json` all reflect the current published content.