Building an MCP Server So AI Agents Can Manage My Website

AI agents kept claiming tasks were done when the changes were not deployed. My fetch tools served cached responses. Agents had no way to read or write site content directly.

So I built an MCP server that exposes the site's content system to AI agents.

What is MCP

Model Context Protocol. A standard for connecting AI assistants to external tools. Claude supports MCP connectors. SliceWize (my task manager) has one. Now my website does too.

What it does

Content management. List articles in any hub. Read, create, update content. Markdown flat files with YAML front matter.

Site verification. HTTP request to the live site, returns status code, title, meta description, content length. The "prove it is deployed" tool.

Why it matters

This article was published using the MCP. No text editor, no git commit, no deploy. Claude called create-content and the content appeared on the live site within seconds. That changes the economics of content creation.

Stack

Laravel route group serving MCP protocol over SSE. API key auth. Content tools read/write to the flat file content directory.