Frequently Asked Questions

Yes. Fabrement generates native Gutenberg blocks, headers and footers that sit alongside whatever theme you already run. It does not replace your theme and it does not add a page builder runtime.

The plugin is free. Over MCP, your agent runs on the AI subscription you already have, so nothing is billed by us. The chat inside wp-admin starts with $3 of free credit, enough to build three or four blocks, and then uses your own OpenAI, Anthropic Claude or Google Gemini key.

No. Describe what you want, or point Fabrement at a Figma frame or a screenshot, then edit the result in the standard WordPress editor. If you do write code, every generated block is open for you to change by hand.

That depends on the route. Over MCP, you connect the agent you already use: ChatGPT, Claude (claude.ai and Claude Code) and Codex sign in with OAuth, and Cursor or any other MCP client connects with a WordPress application password. The agent works on your own plan with that provider, and Fabrement charges nothing for it. Through the admin chat inside wp-admin you can use OpenAI, Anthropic Claude or Google Gemini: new accounts start with free credits, then the chat uses your own API key, billed by that provider.

Yes — on as many sites as you need, including client sites. There is no per-site licence and no limits.

We release updates regularly to ensure compatibility, security, and to introduce new features. Major updates are typically delivered monthly.

Not yet, but it is on its way. Fabrement was submitted to the WordPress.org plugin directory on 30 September 2026 and is in review. Until it is listed, download the free .zip from fabrement.com and install it through Plugins → Add New → Upload Plugin. Once it is approved, you will be able to install and update it from Plugins → Add New like any other plugin.

Yes. Custom connectors work on Claude Free, Pro, Max, Team and Enterprise. The Free plan is limited to one custom connector, and on Team and Enterprise only an Owner can add one. Claude Code is different: it needs a paid Claude plan or a Claude Console account.

The build, yes. Give the agent the design in the evening, and it builds the sections, templates and pages while you sleep. Our own front page went from a Figma frame to WordPress in under ten minutes. What takes longer is the part that needs a person: checking every page and asking for the changes you want. Plan one night for the build and a few days for the review.

No. Nothing is converted line by line. The agent rebuilds each section as a native Gutenberg block, from a Figma file or from screenshots of the current pages, and fills it with your existing text and images. The new pages no longer depend on the page builder, so it can be removed once they are live.

Yes. Give the agent screenshots of the current pages, and it rebuilds the same design on native blocks and one design system. The site looks the same to visitors, but loads faster and is easier to edit.

Not if the addresses stay the same. Keep the page slugs, let the agent set the Yoast titles and descriptions for every page, and redirect any address that changes. Faster pages usually help rankings rather than hurt them.

On a copy is safer. Install Fabrement on a staging copy, rebuild there, review, and then switch. On the live site it also works, because every new page starts as a draft and nothing is published until you say so.

Freelancers and agencies with client sites on old themes or page builders, and site owners who are comfortable working with an AI agent. The Fabrement plugin is free. The agent runs on your own Claude, ChatGPT, Codex or Cursor plan.

No. Fabrement generates clean, semantic code that loads only on the pages using it — there is no shared page builder runtime shipped to every visitor. That is why sites built this way hold their Core Web Vitals.

Yes, completely. Content, structure and styles are all editable in the standard WordPress editor, and the block's code is there if you want to change it by hand.

The blocks belong to you, and you can download the generated code and reuse it anywhere, including in your own theme. What is not automated yet is the moving: today that is a manual step, and while the plugin is installed it is what renders those blocks on your pages. An automated export to your theme is planned.

Your site keeps working. The blocks that already exist stay on the site and keep rendering. An active account is needed to build new blocks or change existing ones, because block code is checked and stored by the service before it is saved.

Yes. Each section is a registered Gutenberg block with its own block.json, server-rendered PHP and scoped CSS. It appears in the block inserter like any other block.

No. It becomes a custom block with editable fields for its content, and its code can be changed by the agent or by hand.

Yes. It is plain WordPress PHP, CSS and JavaScript, and it is written to be what another developer would want to inherit.

Yes. Fabrement turns your site into an MCP server. In Fabrement → MCP Connection you pick your client and follow the setup shown there: ChatGPT, Claude (claude.ai and Claude Code) and Codex connect with OAuth, which needs HTTPS and an administrator's approval; Cursor and any other MCP client connect with a WordPress application password. It needs WordPress 6.9 or newer, and only administrators can connect an agent.

If you would rather not use an agent at all, OpenAI, Anthropic Claude and Google Gemini also work inside wp-admin, through the plugin's own chat, which starts with free credits.

MCP stands for Model Context Protocol. It is an open standard that lets AI applications connect to external tools and data sources through one common interface, rather than through a custom integration for every model and every app.

Not any more. ChatGPT and claude.ai connect with OAuth: you approve the connection once and then work in the chat, with no command line involved. Claude Code, Codex and Cursor take a command or a configuration entry, and the MCP Connection screen generates it for you to paste. If you would rather not use an outside agent at all, the admin chat inside WordPress builds blocks too, without leaving wp-admin.

Treat the connection like an administrator: whether it uses OAuth or an application password, the agent acts as your administrator user, with write and delete access to content. Connect only clients you trust, work on staging when the change is significant, keep versions so a change can be rolled back, disconnect the agent in the Connection access card when the work is done, and review anything that touches payments, authentication or personal data yourself.

A page builder's AI feature produces content inside that builder's runtime, which every visitor then downloads. An agent connected over MCP produces native Gutenberg blocks and PHP that load only on the pages that use them, so the weight your visitors carry is the page they asked for, not the builder.

ChatGPT, Claude (claude.ai and Claude Code), Codex and Cursor. In Fabrement → MCP Connection you pick your client: ChatGPT, Claude and Codex connect with OAuth, Cursor and any other MCP client with a WordPress application password, and the screen generates the setup to paste. Step by step: ChatGPT and Codex, Claude. Rather not use an MCP client? Take the admin chat route instead: OpenAI, Claude and Gemini work there, inside wp-admin.

Yes, for content and structure. With an agent connected over MCP you can create pages and posts, edit any text or image in a block, change menu items, set SEO fields and build new sections, all from the chat you already use. Some things still belong in wp-admin: updating plugins and core, user accounts, and anything a plugin exposes only through its own screens.

Not for everyday changes: text, images, a new page built from sections that already exist. You need one when the system itself changes, when a new kind of section has to be designed, or when something breaks in a way that needs reading the code. The lookup work went away; the judgement did not.

It works through a fixed set of registered operations, not a shell, so there is no file access beyond the blocks it writes and the database operation is read-only. Publishing is a separate step, so a page cannot go public as a side effect of saving it. It does not touch WooCommerce, carts or payments, and it has no role-scoped access: an agent gets the whole surface or none of it.

Through the Abilities API, which shipped in WordPress 6.9 and lets a plugin register an operation with its input, output and permission check, and through an MCP server that exposes those operations to the agent. The agent authenticates as a WordPress user, so the key it holds carries that user's rights: revoke it when the work is finished and keep a backup before large changes.

Partly. WordPress 6.9 added the Abilities API, the standard way for plugins to declare what an AI agent may do, and the official MCP Adapter, a separate WordPress project, turns those abilities into an MCP connection. Out of the box the adapter gives the agent three tools to discover, inspect and run abilities, and abilities stay private until a plugin makes them public. To build blocks, templates or pages, you still need a plugin that provides those operations.

No. You sign in on your own site and approve access, and your AI client (ChatGPT, Claude or Codex) receives a token, not your password. The token only works through the Fabrement MCP server on that site: it cannot sign in to wp-admin or call other REST API routes. It renews itself in the background and stays active until you disconnect it in the Connection access card on Fabrement → MCP Connection, or turn OAuth off.

No. Codex now runs inside the ChatGPT desktop app, but Fabrement treats ChatGPT and Codex as separate clients: each one has its own card in Fabrement → MCP Connection and its own approval. Connect the one you use, or both, and each appears on its own line in the Connection access card.

The application password is sent as a Basic Authorization header. Codex's Bearer token field sends the key in a different format, which WordPress rejects with a 401 error. Put the Authorization header in the header field and leave the Bearer token field empty. With OAuth there is nothing to paste at all.

Yes. Connect each one separately, with OAuth or its own application password. Each connection gets its own line in the Connection access card on Fabrement → MCP Connection, so you can see which one was used last and disconnect one without cutting off the other.

A paid plan. Whether you also need to turn on Developer mode depends on the plan: if Create app already shows on ChatGPT's Plugins page, you do not. On a workspace, an administrator may need to grant permission first. Codex is included in ChatGPT plans and can also run on an OpenAI API key.

Not at the moment. MCP apps created in Developer mode work in ChatGPT on the web, not in the mobile app, and ChatGPT's agent mode does not use them. Work from ChatGPT in a browser, or connect Claude, whose connector is also available in its mobile app.

Yes. You save a Figma token once in Fabrement → API Connection → Figma, then paste a link to the frame you want built. Any Figma plan works, including the free one. Very large frames can fail to fetch: split those into a hero, a feature grid and a footer and build them one at a time, which gives you reusable blocks anyway.

Yes. Every Figma frame becomes native Gutenberg blocks with editable fields, built with real flex and grid layout. Not a page builder layout, and not a screenshot in a container.

Yes. Any Figma plan works, including the free one. Create a personal access token in Figma (Settings, then Security, then Personal access tokens) with read access to file content, and save it once in Fabrement → API Connection → Figma. The token's account must be able to open the file, otherwise the import fails even though the link opens in your browser.

Yes, on a fresh WordPress install with Fabrement connected to your AI agent and a Figma token saved. The agent builds the sections first, then assembles the pages from them, so a whole site fits into one long session. Two conditions: your AI plan needs enough usage for a session that long, because a basic plan can hit its limit halfway, and you wake up to a pre-production site, not a finished one. Expect to review it, fix small layout details and decide what goes live. How close it gets depends on how the Figma file is built.

Build it the way code is structured. Use auto layout on every screen, and frames instead of groups: auto layout converts into real flex and grid, while a group imports with a transparent background. Bind colours, spacing, radii and type to variables and styles instead of typing raw values. Make repeated elements such as buttons, cards, inputs and navigation into components with variants for their states, and name layers clearly, like Header, Card/Price or Button/Primary. Set how each element resizes (fill, hug or fixed), keep icons as vector components, and delete hidden or unused layers. One desktop frame is enough. Add a mobile frame only when you want something specific on mobile, because without one the agent decides the mobile layout itself. A messy file still builds, but the agent has to guess more and you fix more in review.

It gets you most of the way there and keeps your colours, typography and spacing. Complex layouts usually need a round of adjustment, which you do in the WordPress editor or by sending the AI a follow-up instruction. It replaces the tedious part, not your judgement.

Yes. A frame too large to read at once is built section by section: the agent fetches it in parts and assembles the result. Splitting a very long page into separate frames, such as a hero, a feature grid and a footer, still helps, because each one becomes a reusable block instead of one enormous section.

Not to those specific blocks. Fabrement does not fill in a block library like Spectra or Kadence: the agent writes custom native Gutenberg blocks for your design, each with its own fields, so the result follows the Figma frame instead of the closest library block. Pages you already built with Spectra or Kadence keep working next to the new ones.

Most often the site has no HTTPS. OAuth (ChatGPT, Claude and Codex) needs HTTPS, and ChatGPT and claude.ai also need a public site address, so they cannot reach a site on localhost. An application password (Cursor, or the Advanced tab for Claude and Codex) needs HTTPS too: without it WordPress reports «Application passwords are not available for this user on this site». Install an SSL certificate or re-enable application passwords, then try again.

No harm done. An application-password setup is shown once: go back to Fabrement → MCP Connection and generate a new one for your client. The one you did not use can be disconnected in the Connection access card. OAuth connections (ChatGPT, Claude, Codex) have nothing to copy, so this does not apply to them.

Start a new chat: the tool list is read when a conversation begins, so an older chat will not see it. If that does not help, check the WordPress version: the MCP server only starts on 6.9 and newer. On older versions the card is visible in wp-admin while the server never answers.

The setup is incomplete or the password has been revoked. Check that all of it was pasted, since it is long and copying often truncates it, and that the connection is still listed in the Connection access card on Fabrement → MCP Connection. Generating a fresh setup is usually the quickest fix.

Either no key has been added yet, or the free starter credit is spent. Open Fabrement → API Connection, pick Gemini, Anthropic or OpenAI and paste your key, then send your message again. If you see «You have reached your free limit», the starter credits are used up and your own key is now required.

Claude connects to custom connectors from Anthropic’s cloud, not from your computer, so a site on localhost or a private network is out of its reach. For a local site, use Claude Code, which connects from your own machine: with OAuth when the site has a trusted HTTPS certificate, or with an application password.

For building and managing a native Gutenberg site, yes. For operating a site built on Elementor, or for running WooCommerce, WPVibe covers ground Fabrement does not.

WPVibe. Fabrement builds native Gutenberg blocks and does not edit page-builder layouts.

Both plugins are free. WPVibe meters WordPress actions per day on its service, from 100 a day on the free plan. Fabrement does not charge per action over MCP; you pay only for your own AI client.

WPVibe has an approval panel with dry-run previews and an audit log today. Fabrement gives the agent no file access beyond the blocks it writes, a read-only database operation and publishing as a separate step, but has no approval panel yet.

Not directly. The Adapter is the bridge that exposes abilities registered by core and plugins. Fabrement is a plugin whose operations build and run the site. Without plugins that register building abilities, the Adapter gives an agent little to build with.

Not by itself. Its default server offers tools to discover, inspect and run abilities. Building blocks needs a plugin that registers that ability.

They control different things. The Adapter keeps abilities private by default and checks permission for each one. Fabrement checks the code an agent writes before it is saved and keeps publishing separate, but exposes its operations all or none. In both cases the agent acts as a WordPress user, so give it only the access the job needs.

Not in 7.2. The roadmap plans to publish it in the plugin directory and let administrators enable MCP from the AI plugin. AI features have to show adoption and practical value before they are considered for core.

For building on a live site on any host, yes. For local development with files, WP-CLI and screenshots, Studio covers ground Fabrement does not.

Its MCP server works with local Studio sites. Studio Code, its AI assistant, can also work on WordPress.com and Jetpack-connected sites. For other hosts you deploy the work yourself.

Studio keeps mistakes on your computer until you sync them. Fabrement works on the site, but through fixed operations: block code is checked before it is saved, the agent has no file access beyond the blocks it writes, and publishing is a separate step.

Both start free. Studio is free and open source, and Studio Code is free during its beta within a monthly limit. The Fabrement plugin is free and does not charge per action over MCP; you pay only for your own AI client.

It is scheduled for 8 to 10 December 2026. Beta 1 is planned for 20 to 22 October and Release Candidate 1 for 17 to 19 November.

Not in core. The roadmap puts the MCP work in the AI plugin: the MCP Adapter published in the plugin directory and enabled from the AI plugin, with the administrator choosing which abilities are exposed. AI features have to show adoption and practical value before they are considered for core.

It already can, through plugins that register abilities or run their own MCP server. 7.2 makes connecting easier and explores write abilities in the AI plugin; it does not promise a full set of write abilities in core.

Not as a finished feature. Agent identity and delegation is on the roadmap as a model to develop: auditable identities, appropriate permissions, and access managed apart from human accounts. Until it ships, an agent acts as the WordPress user whose application password it holds.

Yes, if you want the agent to build and not only edit. New blocks written as code, a shared design system, templates and archive pages are outside the 7.2 roadmap.