The Model Context Protocol (MCP) is an open protocol for connecting AI applications to external tools and data sources in a standard way.
The Problem It Solves
Every AI assistant needs integrations — files, databases, ticketing systems, APIs. Without a standard, each assistant and each tool needs custom glue code. MCP defines a common interface so a tool implemented once can be used by any compatible application.
The Pieces
- MCP servers expose capabilities: tools the model can call, resources it can read, and prompts it can use.
- MCP clients live inside AI applications and connect to servers.
- The application decides which servers are connected and presents their tools to the model.
What It Enables
An assistant can search your documents, query a database, create a ticket or read a repository through servers you choose to connect, without the application needing bespoke code for each.
Security Considerations
- Trust: only connect servers from sources you trust; a malicious server can expose harmful tools or return manipulative content.
- Permissions: grant servers the least access needed; prefer read-only where possible.
- Prompt injection: content returned by tools can contain hidden instructions.
- Approval: require confirmation for sensitive actions.
- Data exposure: understand what data flows to the model and its provider.
Getting Started
Many AI applications and developer tools support MCP, and there are open-source servers for common systems. Start with a single read-only server and expand as you understand the behaviour.