What is the Model Context Protocol (MCP)?

What problem does MCP solve?

Before MCP, every AI application needed its own integration for every tool or data source it used. Ten applications and ten systems meant up to a hundred bespoke connectors, each with its own authentication, error handling and drift. MCP replaces that with one protocol: a system is exposed once as an MCP server, and any compatible AI client can use it.

How does MCP work?

MCP uses a client–server model. An AI application (the host) runs an MCP client that connects to one or more MCP servers. Each server describes what it offers in a structured, machine-readable way:

• Tools are actions the model can call, such as creating a record or running a search.
• Resources are data the application can read into context, such as files or records.
• Prompts are reusable, parameterised instructions for common tasks.

Servers can run locally alongside the application or remotely over HTTP. Remote servers use standard OAuth-based authorisation, so access can be scoped to a user and revoked like any other integration.

Who created MCP?

Anthropic introduced the Model Context Protocol as an open standard in November 2024. It has since been adopted across major AI platforms, developer tools and enterprise software, which is what makes it useful: one server works with many clients.

When should a product use MCP?

MCP makes sense when you want AI systems — your own or your customers’ — to act on your product’s data and workflows through a governed interface. Typical cases:

• Making a product usable from AI assistants your customers already use.
• Giving an internal agent controlled access to several systems at once.
• Replacing one-off model integrations that are hard to secure and maintain.

It is less useful when a single application calls a single API in a fixed way. A direct integration is simpler there.

What does a good MCP server look like?

The protocol is the easy part. The product decisions sit in the server design: which actions are exposed, how narrowly each tool is scoped, which operations need human confirmation, how errors are reported back to the model, and how every call is logged. A good server exposes fewer, well-described tools with explicit permissions rather than mirroring an entire API.

How Digitalis uses MCP

Digitalis builds MCP-native products in high-consequence domains, where every tool call needs explicit permissions, confirmation for consequential actions and an audit trail. The same discipline applies to any organisation exposing its systems to AI.