How JBiz Media Runs Its Own Website Through an MCP Server

MCP is not a side project at JBiz Media, it is how a lot of our own work gets done. We use it for SEO audits, for building pages in bulk, for AI website development and for our own lead generation. The clearest example is this website, which an AI assistant manages through an MCP server. These are our own operations, so this is a worked example rather than a client case study.

Where we use MCP ourselves

• SEO audits: the assistant pulls Search Console data and runs checks on content, links, redirects, schema, performance and accessibility, then reports what needs fixing
• Mass page builder: new service, industry and blog pages are created from a strict content model, so many pages can be produced consistently and malformed ones are rejected
• AI website development: the assistant reads, drafts, validates and publishes site changes, then confirms the exact build is live
• Lead generation for our own business: MCP tools connect the data and systems behind our outreach so the assistant can research, organize and prepare work for us to review

Having the same pattern behind each of these is the point. Once the tools are narrow and checked, adding another job is a matter of adding another tool, not starting over.

Why we built it this way

Marketing work is repetitive and detail-heavy: new pages, updated copy, fixed links, redirects, audits, lists of prospects. Done by hand, that means opening files, remembering conventions and hoping nothing breaks. We wanted an assistant that could take a plain-language request such as add a page for this service and do it correctly every time, which meant giving it structure rather than freedom.

What the website server lets the assistant do

• Read any page and see its real data before changing it
• Preview a change and see exactly what it would alter before it is written
• Create or update pages using a strict content model, so a malformed page is rejected
• Run quality checks on content, links, redirects, SEO, schema, performance, accessibility and security
• Confirm that a specific build is live before testing it
• Read Google Search Console data to see what people search for and where pages are underperforming
• Roll a change back if a check fails

Where the guardrails are

The most important design choice was limiting what the assistant can touch. Page content goes through typed tools that validate every field. Navigation is generated from content instead of hand-edited. Anything destructive, such as renaming a page or bulk-changing text, requires an explicit confirmation and a preview first. Page text, search data and fetched content are treated as untrusted, so instructions hidden inside them are ignored.

That is the same pattern we apply when we build servers for other businesses: narrow tools, previews before writes, logs of every action, and a human decision on anything that cannot easily be undone. Our MCP development services are built around it.

What it changed day to day

Publishing a page now has a repeatable path: read a similar page, preview, write, link it into related pages, confirm the build, run checks. Mistakes tend to be caught by a gate instead of by a visitor. Writing the content still takes judgment, and we review what the assistant produces. The server removes the clerical risk, not the editorial responsibility.

What we got wrong along the way

• Early checks tested a protected preview link and reported false failures; we now test the production domain
• Some of our own rules flag wording even when it is negated, so we learned to phrase claims the way the checks expect
• Bulk operations are powerful enough to need a plan step every time, because a wrong find-and-replace spreads quietly

None of this is exotic. It is ordinary engineering discipline applied to AI access. The same pattern extends to any marketing work that lives in several tools, such as pulling search and campaign data into one place for an assistant to analyze, which is how we think about AI-powered SEO too.

Frequently Asked Questions

Does the AI publish without anyone checking?

Changes can go live quickly, so we rely on previews, validation gates and post-deploy checks, and we review the content itself. Sensitive actions need explicit confirmation.

Could a business like mine do this?

Yes, if your site or systems have a structured way to read and change content or data. The work is designing the tools and limits, which is what an MCP build covers.

Is this a client case study?

No. It is our own operations. We say so deliberately: it shows how we build and use MCP servers, not results achieved for a customer.

What is MCP, briefly?

An open standard that lets AI assistants use tools and data through one interface. Our guide to what MCP is explains it in plain English.