CMS for AI Agents: WebMCP Built Into AliothPress - AliothPress

EN

DE

ES

AliothPress

Home

Features

Explore CMS features

Plugins

Installation Guide

Documentation

Content Management

Multilingual System

Page Builder

AI Assistant

Media Library

File Manager

Form Builder

Newsletter

SEO, AEO & Performance

Themes

Menus

User Management

Security

Backup & Restore

Favicon Generator

SSL / HTTPS

Licensing & Updates

Privacy & GDPR

Accessibility

Tech Stack

Download CMS

Demo

AliothPress: a CMS for AI Agents (WebMCP Built In)

· CMS AliothPress

AliothPress is a self-hosted CMS with native support for AI agents through WebMCP. Any site built on it can let AI agents search its content, describe and submit its forms, and, inside the admin panel, create posts, build pages, upload images, and translate content. The feature is off by default and controlled by two switches on the AI Assistant page in the admin panel. AliothPress never enables AI agent access automatically. Site owners explicitly decide whether to expose public tools, admin tools, both, or neither.

What is WebMCP

WebMCP is an emerging web standard that lets websites offer tools to AI agents, the same way robots.txt offers rules to search engines. A site declares what an agent is allowed to do. The agent discovers it automatically.

Discovery : every AliothPress site answers at /.well-known/webmcp with a machine-readable manifest

In-page registration : tools register via document.modelContext / navigator.modelContext

Optional polyfill : a compatibility polyfill adds the API on browsers without native support, switched on with one checkbox on the AI Assistant page. The origin trial token is pasted on the same page

WebMCP support ships in the CMS core, ready to use as installed, and relies on zero plugins or third-party services.

What AI agents can do on your site

With the public surface enabled, an agent visiting any page of your site can:

Search your content : search_site queries published posts and pages

Read the site overview : auto-generated /llms.txt summarizes the site for AI, grouped by language

Understand your forms : describe_form returns fields, types, and requirements

Submit forms for the visitor : booking, contact, signup. The visitor asks, the agent fills and submits the form

Example: a visitor tells the AI agent in their browser "book me into the Saturday workshop". The agent finds the form on your page, fills it, and submits it. Your honeypot spam protection and rate limits still apply.

What AI agents can do in your admin

With the admin surface enabled, an agent working inside your authenticated admin session gets up to 24 tools:

Content : generate posts and pages, set their slug, SEO, social card, FAQ, and Schema.org metadata in the same call, create language versions linked into one translation group, translate text, optimize SEO metadata, and edit existing posts and pages field by field: only the fields the agent names change, everything else is preserved

Media : upload images into the media library, one at a time or as a whole set under a single approval (the same pipeline as human uploads: optimization, WebP/AVIF variants, EXIF stripping, mandatory alt text), and attach them as the featured, Open Graph, or Twitter image in one step

Page building : create Page Builder pages singly, in batches, or as translations, read a machine-readable reference of every block type, and set or update page blocks

Bulk work : create dozens of posts or pages in one go with a single approval dialog, Page Builder layouts and image sets included, ideal for migrations and documentation runs

Site management : build menus, check slugs, list and search content, navigate admin sections

The agent acts as the signed-in user and reuses the exact routes humans use: same permissions, same validation, same sanitization, same audit log. A separate agent backdoor simply does not exist in the codebase.

One thing an agent can never do is publish. Agents create and edit drafts. A request to publish is demoted to a draft server-side, independently of what the agent or its tools claim. The rule is enforced twice, in the tool layer and on the server, so publishing stays a human decision, made in the admin UI.

Approvals that scale with the job

Every write asks the signed-in user for confirmation. For one post, that is one dialog. For bulk work, the dialogs scale instead of multiplying:

Batch approval : when an agent creates many posts or pages at once, or uploads a set of images, a single dialog lists every item with checkboxes: select all, or approve a hand-picked subset. Skipped items are reported back to the agent, so it knows exactly what happened to each one

Autopilot : any approval dialog can enable time-boxed auto-approval: 15 or 60 minutes, in that browser tab only. A visible badge counts down and revokes with one click. Only a human can switch it on. It does not exist as an agent tool. And even under autopilot, publishing...