The Model Context Protocol

Your AI's Bridge to the Real World

I’m a bit late to the game, but over the past few weeks, I’ve been installing and using Model Context Protocol (MCP) and can see how this changes what and how you use AI tools.

Let’s get into it …

I think we all agree that AI assistants are brilliant at answering questions and writing content and code, but they've always had one frustrating limitation: they're stuck in their training data bubble. Want your AI to check your website analytics? Copy and paste. Need it to publish a blog post? More copying and pasting. Want it to access your company database? You're out of luck entirely.

MCP changes all of this. Think of it as giving your AI assistant a direct phone line to your real-world tools and data, transforming it from a clever chatbot into a capable digital colleague.

What MCP Actually Does

The Model Context Protocol is an open standard that lets AI models connect directly to external data sources and tools. Rather than being limited to what they learned during training, AI assistants can now access live information from your databases, APIs, files and web services.

Practically, what does this mean? Instead of asking Claude to write a blog post that you then manually copy into WordPress, you can tell Claude to “write this article and publish it to my blog on Tuesday at 9am”. The AI handles everything from content creation to scheduling, working directly within your systems.

MCP solves what developers call the “NxM problem”. Previously, each AI model needed custom integration code for every tool or service it wanted to connect with. If you had 10 AI models and 10 services, you'd need 100 different integrations. MCP creates a universal connector, like USB-C for AI, where each service implements MCP once and works with any compatible AI system.

The Technical Foundation (Without the Jargon)

MCP uses a client-server architecture that's straightforward in practice. The MCP server sits between your AI assistant and your data source (whether that's a database, file system or web service). The AI assistant contains an MCP client that discovers and connects to these servers automatically.

When you ask your AI to fetch information or perform an action, the client sends a standardised request to the appropriate server, which then interacts with your actual systems and sends back the results. This happens in the background whilst you continue your conversation.

The protocol uses standard communication methods like JSON-RPC over HTTP, meaning it works reliably across different systems and networks. Most importantly, MCP respects your existing security settings. If your user account can't access certain data, neither can the AI working on your behalf.

Getting started with MCP isn't as intimidating as it might sound, though it does require some technical comfort. Most implementations involve downloading an MCP server for your specific service (many are available on GitHub), configuring authentication and adding the server details to your AI application's settings.

For common services like GitHub, Google Drive or PostgreSQL databases, you'll often find prebuilt MCP servers. These handle the technical complexity of connecting to APIs and databases, leaving you to focus on configuration rather than coding.

The general process involves installing the MCP server (usually through npm or pip), setting up authentication credentials (API keys or service accounts) and then registering the server with your AI application. Most modern AI tools that support MCP (I use Claude and Cursor primarily) will automatically discover and connect to configured servers.

WordPress MCP

I've been running an educational blog for teachers for several years, and my publishing workflow was becoming increasingly cumbersome. I'd write content with AI assistance, then spend ages copying, formatting, categorising and scheduling posts manually. The constant switching between applications was breaking my flow and eating into time I'd rather spend on content strategy.

Setting up the WordPress MCP made it so much easier. Now I can tell Claude to “write an article about classroom management techniques, schedule it for next Tuesday morning, and add it to the teaching strategies category”. Obviously, my prompts and templates are a bit more sophisticated, but you get the idea. Claude writes the content, formats it properly with headings and structure, assigns the right categories and tags, and schedules publication automatically.

The WordPress MCP server connects through your site's REST API, which means it works within WordPress's existing security framework. Claude operates under the same permissions as my user account, so there's no additional security risk.

Beyond publishing, Claude can now audit my existing content for consistency, update site settings, manage media files and even analyse posting patterns to suggest optimal publication schedules. It has changed from a writing assistant into a comprehensive website manager.

Google Analytics MCP

My second MCP implementation connected Google Analytics to my code editor, Cursor, and AI assistant Claude. I was constantly switching between coding and checking website performance data, trying to understand which optimisations were actually working. The Google Analytics interface has always felt overwhelming to me, and I often struggled to find simple answers to basic questions about traffic and user behaviour.

Setting up the Google Analytics MCP required creating a service account in Google Cloud, enabling the necessary APIs and downloading authentication credentials. The trickiest part was understanding the difference between the Property ID (which MCP needs) and the Measurement ID (which won't work). Once configured, though, the integration worked well.

Now I can ask my AI assistants questions like "which pages had the highest bounce rate yesterday?" or "how's my traffic trending this week?" without leaving my editor. The AI can analyse traffic patterns, suggest content improvements based on user behaviour and even help identify technical issues affecting site performance.

The natural language interface means I can explore my analytics data conversationally rather than hunting through GA4's complex interface. Questions like "show me the top performing blog posts from last month" get instant, formatted responses that help inform my content strategy decisions.

Using the Google Analytics and WordPress MCPs from Claude.

Broader Implications

MCP represents a fundamental shift in how we interact with AI systems. Rather than AI assistants being isolated tools that require constant input and output management, they become integrated parts of our digital workflows. This transformation aligns with Andrej Karpathy's observation that LLMs are becoming the new operating system for knowledge work.

The protocol's open standard nature means we're not locked into proprietary ecosystems. As more services implement MCP servers, AI assistants become increasingly capable without requiring separate integrations for each new tool. This standardisation benefits both users and service providers, creating a more connected and efficient digital workspace.

What excites me most about MCP is how it makes AI assistance more contextual and actionable. Instead of general advice based on training data, AI can now provide specific recommendations based on your actual data, execute real actions in your systems, and maintain context across different aspects of your work.

The setup process still requires some technical knowledge, but as MCP adoption grows, we'll likely see more user-friendly configuration tools and broader support across different platforms. For now, the investment in learning MCP pays dividends in workflow efficiency and AI capability.

Until next time, Niall