Bob Browning's blog

MCP (Model Context Protocol) solves a fundamental problem: how do you give an AI model access to external data like databases, files, or APIs without embedding everything in each request? Introduced by Anthropic in November 2024, MCP is now mature enough (15+ months old) for serious production use. Think of it like a USB-C port for AI—a standardized way to connect AI models to external systems.

How MCP Works

Instead of stuffing all your data into every API call, you connect the AI to an MCP server that provides access to your data on-demand. The AI can then query only what it needs, when it needs it.

MCP servers can expose:

• Databases
• File systems
• APIs (search engines, calculators, etc.)
• Business tools (Google Drive, Slack, etc.)

You Can Use MCP Right Now

In Claude Desktop or the web app, click the plus sign next to the input box and select “Connectors.” These are pre-built MCP integrations including:

• Google Drive search
• Google Calendar
• Notion
• And many others

My Use Case: Fintech Feature Matrix

At FintechBenchmark.com, we needed to create a features matrix for ~4,000 fintech products. Each category has ~100 features that need to be matched to products.

The old approach: Send massive amounts of data with every API call—expensive, slow, and limited by context windows.

The MCP approach: Set up an MCP server that lets Claude query our database on-demand, retrieving only relevant products and features as needed.

MCP Support Across Providers

• Anthropic (Claude): Full MCP support via Claude Desktop and API (both local and remote servers)
• OpenAI: Partial MCP support via their Responses API (remote servers only, tools-only—no resources/prompts yet)
• Google (Gemini): MCP has been adopted by Google as a standard for connecting AI agents to external data and tools. As of early 2026, Gemini supports MCP through several primary channels:

Why I’m Excited: Data Privacy

Our feature matrix will become a unique, proprietary dataset. We need to:

✅ Allow AI queries for logged-in users on our site
❌ Prevent this data from being used for AI training

The crucial question: When an AI model accesses data through MCP, is it protected from training?

The answer: Yes—both Anthropic and OpenAI state that:

• Commercial API data (including MCP tool calls and responses) is NOT used for training by default
• MCP server content is explicitly excluded from training data
• This protection applies even if you submit feedback via thumbs up/down buttons

Source: Anthropic’s Privacy Policy

The Bottom Line

MCP enables powerful, data-driven AI applications while maintaining data privacy. For businesses with proprietary data, this is game-changing.

That said: As with everything in AI, verify the terms yourself, especially for mission-critical applications. Privacy policies can change, and what’s true today may not be true tomorrow.