The MCP-first Agentic CMS

The MCP-first Agentic CMS
We've built an MCP-first content management system to make website content editing easy and cheaper. If, like us, you spend more time in an agent interface such as ChatGPT or Claude, and you own a website, you would probably feel at home getting rid of your CMS user interface and editing the site from the app you already use.

Introducing the MCP-only CMS

At Launch Lab we’ve built hundreds of websites over the 11 years we’ve been in business. Almost all of them have a CMS, and for most of them we’ve trained our clients on how to use that CMS.

Our current preferences are Webflow and certain headless CMSs. Over the years we’ve also used Contentful, Netlify CMS, Django CMS, FeinCMS, Storyblok, Strapi and, very early on, WordPress.

The one common thread is that they all have a user interface and a learning curve.

What if there wasn’t a UI? What if you could edit your website using the Claude or ChatGPT subscription you already pay for?

This isn’t a thought experiment. We’ve built exactly this: an AI-native CMS you use from inside ChatGPT or Claude.

There is no CMS dashboard. Your team adds one secure connector to the AI assistant they already use, the same way IT adds any approved workplace tool. Then, in a normal chat:

You: Swap the headshot on the About page for the attached photo, and update Jane's title to Chief Operating Officer. Or, create a new landing page for a Google Ad campaign using the attached copy. Be sure to use our lead-gen page template for the layout.

Assistant: Done in draft. Here's a before / after and a preview link. Want me to publish?

You: Yes

What is MCP?

MCP (Model Context Protocol) is an open standard that gives AI assistants a safe, well-defined way to connect to outside tools and data. Anthropic, the maker of Claude, released it in late 2024, and it has since been adopted across the industry, including by OpenAI, Google and Microsoft.

Think of it as the USB-C of AI: one standard plug, so any assistant can connect to any service that offers it.

When you add a connector to ChatGPT or Claude, for Google Drive, for Slack, or for your website, that’s MCP under the hood. A connector does not hand the AI the keys to everything. It exposes a specific menu of actions the service chooses to allow, and the service itself checks every request against who you are and what you’re permitted to do.

That’s what makes it fit for business use. IT approves the connector once, it works for the whole team, and the AI can only ever do what the connector permits.

What do the current CMS providers offer?

Most of them also provide an MCP server, including Webflow, Sanity and Storyblok.

So yes, if you already run one of these platforms, you can now talk to it. But the MCP is a new door into the same building. You’re still paying for the platform and administering its seats, and most of these MCPs were built for web developers, with no guardrails stopping a well-meaning prompt from changing things that shouldn’t change.

Webflow has also announced Source, an agentic workspace where AI agents work on your site alongside your team. The industry’s biggest visual builder is betting that the next website interface is an agent, not a dashboard.

Why did we build an MCP-first CMS?

Considering the established players already let you manage website content via an MCP, why would we build our own agentic CMS?

You could say those platforms are becoming MCP-first. We’ve gone further and built an MCP-only CMS, which has some key advantages:

  • It's cheaper to run. There is no user interface to look after, so it costs less for the website owner to use.
  • There's a lower learning curve. Set it up once and your team uses the agent interface they already know. There is no new UI to learn.
  • It saves time. Agent rules and skills are built in, so you don't have to write them. If you're unsure what those concepts are, think of them as a harness for your agent. A harness that only permits improvements to your website. Never go backwards and never break it. Never leave stale JSON-LD schema (structured data). Never struggle to add a new page. Just prompt.
  • An agentic CMS made for content editors. Other CMS MCP servers are made for developers. This is made for content editors.

Does this signal the end of the established CMS companies?

Definitely not. There will always be reasons to choose the right tool for the right job. Tools like Webflow offer much more than a CMS, and some projects will need those additional features.

But for the enormous middle, the marketing sites, the professional-services firms, the companies whose website changes a few times a month and who mostly want it fast, secure, findable by AI, and editable without a support ticket, we think the era of learning a CMS is ending.

For those who need to keep costs low, want simplicity, and still want the power of a good content management system, we think you’ll love what we’ve created.

Want to be one of the first to use it?

Register your interest in our MCP CMS. If you’re weighing up Webflow, a headless CMS, or another approach for a website redesign, talk to us before you decide and we’ll help guide you to the right solution for your project.

Let's chat

Get an obligation free web design or web development quote from an experienced local Australian development team.