How to build a WordPress site with AI and MCP

AI can now do more than draft WordPress copy. With the Model Context Protocol (MCP), a compatible agent can read a site’s current structure and call specific tools to create pages, templates, navigation and metadata. The important part is the workflow: inspect first, define the system, build through stable records, then verify what is live.

What is MCP in WordPress?

MCP is a standard way for an AI client to discover and use documented tools supplied by another system. In Shilpi, WordPress hosts the MCP server. The theme decides which website operations exist and whether the server is enabled. This gives the agent a narrow, understandable surface instead of general server access.

Step 1: write a structured website brief

State the audience, business goal, required pages, content source, visual direction and success action. Include non-negotiables such as responsive behaviour, keyboard access, meaningful alt text and restrained scripts. Shilpi’s website prompt builder turns these decisions into a reusable brief.

Step 2: connect an MCP-compatible AI client

Enable the server in Shilpi and use OAuth where possible. OAuth lets the client sign in through the browser without storing a WordPress password in a configuration file. Application passwords remain available for clients that need them and can be revoked individually.

Step 3: make the agent inspect before writing

A safe first run reads the site identity, WordPress and theme versions, existing pages, templates, variables, classes, menus and media. This prevents duplicate pages, parallel design systems and accidental replacement of work that already exists.

Step 4: establish design tokens and reusable structure

Colours, typography, spacing and breakpoints should be decisions, not scattered values. Define variables first, then let page styles refer to them. Build the site-wide header and footer once. Use Parts for reusable fragments and templates for dynamic routes.

Step 5: create pages as semantic HTML

Each page needs one clear purpose, one h1, a logical heading order and useful internal links. Shilpi stores body-only HTML as the page itself, with presentation in page CSS and behaviour in page JavaScript. The result remains readable even when the original agent is no longer involved.

Step 6: use WordPress routing for dynamic content

Build a Single template for posts, an Archive template for categories and dates, a Search template for queries and a 404 template for missing URLs. Query loops can inherit WordPress’s current query so pagination, taxonomy filters and plugin hooks keep working.

Step 7: complete content operations

A finished website is more than its canvas. Create the primary menu, assign categories and tags, upload media to the WordPress library, and write SEO metadata. Shilpi reads and writes through supported SEO plugins so their caches and sitemaps remain correct.

Step 8: verify before reporting completion

• Read the saved records back and resolve warnings.
• Open every public URL and test desktop, tablet and mobile layouts.
• Check navigation, forms, keyboard focus and image alt text.
• Confirm page titles, descriptions, canonical URLs and indexing rules.
• Disable temporary permissions that are no longer required.

Common mistakes

Do not give an agent unrestricted filesystem access for a page-building task. Do not let it invent a second colour system without reading the first. Avoid inline styles, unregistered third-party scripts and hotlinked images. Finally, do not confuse “generated successfully” with “verified publicly.”

Frequently asked questions

Which AI tools can connect to a WordPress MCP server?

Any client that supports MCP can potentially connect. Shilpi is designed for tools such as Claude Code, Cursor and other compatible agents.

Can AI update an existing WordPress website?

Yes. Stable IDs and slugs let an agent read and update existing records instead of creating duplicates.

Is coding knowledge required?

No for routine builds, but understanding HTML structure and WordPress concepts helps when reviewing advanced work. Shilpi’s visual builder remains available for human edits.

Start with the system, not the prompt

Read the Shilpi documentation and review the AI and MCP features. The quality of an AI-built WordPress site depends less on a clever one-line prompt than on clear constraints, structured tools and a disciplined verification loop.