People asking AI assistants about your product or project often get outdated answers. An MCP server gives AI tools direct access to your up-to-date content. Try this approach with my MCP server for guidance on IR writing and security product strategy.
How to make your content, be it a set of articles or project documentation, available to people’s AI agents? They work better when they can access your content through an MCP interface instead of relying on a point-in-time snapshot of the model’s training data or searching the web. For instance, when I work on projects that use Cloudflare, my coding agent knows the latest specs because I added the Cloudflare docs MCP server to its configuration.
Below is a practical approach you can use to make your own site available to AI agents. It uses Cloudflare Workers to run a serverless MCP “server” remotely. The code is available as an open-source project that you can replicate and customize for your needs.
This Approach in Action: My Website MCP
I created an MCP server for my website that uses this approach. When you connect your AI agent to it, you give it access to the following:
- Incident response writing advice: Your AI drafts IR reports from raw notes or gives feedback on existing drafts, applying the template and principles I teach in my cybersecurity writing course.
- Security product strategy framework: Your AI creates, stress-tests, or compares cybersecurity product strategies, using frameworks for positioning, pricing, go-to-market, trust, and defensibility scoring.
- My perspectives on cybersecurity: Your AI draws on the expertise I’ve shared over the years on my blog.
My MCP server shares guidance with your AI for local analysis, so your notes, drafts, and proprietary details stay in your environment.
The easiest way to add my MCP server to your setup is to ask your AI tool to do it. Give it the URL https://website-mcp.zeltser.com/mcp, and your tool will automatically determine when to use the server to assist you.
Why the Model Context Protocol
Web search and retrieval augmented generation (RAG) each put obstacles between AI agents and your content. Web search requires the tool to decide to search, wait, and hope the right results appear. RAG can be tricky to deploy, and readers can’t easily add it to their AI tools.
An MCP server makes your content a native capability of any AI tool that connects to it. The AI discovers and queries your content automatically, in a fast and token-friendly way.
How the Solution Works
The system I built has three components:
- Adapter: A build-time tool that processes your site’s markdown content and generates a search index.
- Cloudflare R2: Hosts the search index so the Worker can retrieve it on demand.
- Cloudflare Worker: Implements the MCP “server” in a serverless manner, handling search and retrieval requests from AI tools.
When AI agents connect, they discover your site’s capabilities and can query your content as needed.
The Cloudflare architecture keeps costs minimal and is often free even at the entry-level pricing tier. I published adapters for generating content for Astro, Hugo, and generic Markdown files. You or your AI coding agent can easily adapt them for other static website generators.
Getting Started
The repository I created includes the Worker code, build-time adapters for generating search indexes, and configuration examples for AI tools such as Claude Desktop and Claude Code. You can clone it and customize it or point your AI agent at it and let it do the work for you.
