Model Context Protocol (MCP)

Model Context Protocol (MCP) is an open standard introduced by Anthropic that provides a universal way for AI models to connect to external tools, data sources, and services. Think of it as a USB standard for AI — just as USB provides a universal way to connect devices to computers, MCP provides a universal way to connect capabilities to AI models.

The problem MCP solves

AI models are powerful but isolated. An LLM can write brilliantly, reason well, and analyse text — but by default it cannot:

• Access your company's database
• Read your emails
• Check your calendar
• Query your CRM
• Run code on your machine
• Search the web in real time

Before MCP, connecting AI to each external tool required custom integration work. Every AI provider, every tool, and every developer had to build bespoke connections. If you wanted Claude to access your Salesforce data and your Google Calendar, you needed two separate custom integrations.

MCP standardises this. A tool that implements MCP can be connected to any AI model that supports MCP — one standard, universal compatibility.

How MCP works

MCP uses a client-server architecture:

1. MCP Server: A lightweight programme that wraps a tool or data source and exposes it through the MCP standard. For example, an MCP server for Google Calendar exposes functions like "list events," "create event," and "check availability."

1. MCP Client: The AI application (like Claude Desktop or Claude Code) that connects to MCP servers. The client discovers what tools are available and can call them when needed.

1. Protocol: The standardised way clients and servers communicate — what functions are available, what inputs they need, what outputs they return.

When you ask Claude "What meetings do I have tomorrow?" and a Google Calendar MCP server is connected, Claude can call the server's "list events" function, get real data, and provide an accurate answer.

What MCP enables

With MCP, AI assistants can:

• Access real-time data: Current stock prices, live weather, your latest sales figures — not just training data
• Take actions: Send emails, create tickets, update databases, schedule meetings
• Use specialised tools: Run code, perform calculations, search databases, query APIs
• Chain capabilities: Combine multiple tools in a single workflow — "Check my calendar, find a free slot, draft an invitation, and send it"

The MCP ecosystem

The MCP ecosystem is growing rapidly:

• Pre-built servers: Servers exist for popular tools — Slack, GitHub, Google Drive, databases, and more
• Custom servers: Organisations can build MCP servers for their internal tools and data sources
• Community contributions: The open-source nature of MCP means developers worldwide are building and sharing servers

MCP vs traditional API integration

The difference is standardisation:

• Traditional integration: Each tool-AI connection is custom. Connecting 10 tools to 3 AI models requires up to 30 custom integrations.
• MCP: Each tool builds one MCP server. Each AI model builds one MCP client. 10 tools + 3 AI models = 10 servers + 3 clients, and everything works together.

Security and control

MCP includes security considerations:

• Permission model: You control which MCP servers are connected and what actions they can take
• Authentication: Servers can require authentication before granting access
• Scope limitations: You can restrict what data and actions are available to the AI
• Logging: All MCP interactions can be logged for audit purposes