The Model Context Protocol (MCP) is an open protocol, initially driven by Anthropic, that defines a standardized way to connect language models with external data sources and tools. Instead of programming custom integrations for each service, MCP establishes a common language between the model (client) and the resources it wants to query or manipulate (servers).
Its importance lies in interoperability: the same MCP server can be used with different models or applications, which reduces duplication of effort and facilitates maintenance. An MCP server typically exposes three types of capabilities:
- Resources: data that the model can read (files, logs, queries).
- Tools: actions it can execute (send emails, query APIs).
- Prompts: reusable templates for specific tasks.
In practice, MCP allows an assistant to access, for example, a code repository or a corporate database without coupling that logic to the model itself, improving modularity and deployment security.