Breaking down MCP

🎯 tldr

• MCP enables Agentic AI at scale

• MCP shifts the burden of maintaining multiple integrations, from the developer down to the service provider

• Both Google and Open AI adopted MCP, meaning the top players in AI are coalescing around this standard, which makes it the de facto standard

Table of Contents

• 🧐 What Happened

• 💡 Why It Matters

• ✅  What It Means for Your Company

• 👀 What We're Watching

🧐 What Happened

In November 2024, Anthropic released the Model Context Protocol (MCP) — a standardized way for AI models to interact with external services and tools. Recall, LLMs can't actually execute anything for you - they just predict the next token in a sequence. BUT when paired with services like Salesforce, Slack, Internet search, etc. they can perform actions. Enter the Agent.

Before MCP:

After MCP:

Before a standardization layer, developers build and maintain fragile, one-off integrations (building API wrapper, handling auth, schema changes, etc), MCP lets service providers expose their tools in a consistent way that LLM models can understand and use. It moves the burden of maintenance and scale from the developer to the service provider.

💡 Why It Matters

Today, most AI agents rely on custom-built API integrations for every tool they interact with. Most of these companies employ teams of engineers to maintain these APIs since they cost a lot of build, are fragile and it's hard to scale.

• MCP changes the game:

  • Offers a shared protocol for LLMs to retrieve context and take actions

  • Reduces engineering lift by moving integration work to tool vendors

  • Accelerates the shift toward Agentic AI — systems that autonomously reason and act across tools

With OpenAI recently announcing support for MCP, it's quickly becoming the standard for agent architectures.

✅  What It Means for Your Company

While the space is still evolving, I recommend asking your vendors and "Agentic AI" providers about their integration approach:

• Are they adopting MCP or another protocol?

• If they're using custom integrations, what's their plan for maintaining these connections at scale? 

• How will they ensure reliability as the services they connect to evolve?

👀 What We're Watching

We're monitoring how the AI integration ecosystem develops around these standards. I'll be attending Data Council next week and will report back on MCP adoption discussions among practitioners. Stay tuned!