inboxmcp
EN DE

← Back to the blog

Basics

The Model Context Protocol, explained

What the Model Context Protocol is, the problem it addresses and the role it plays in connecting AI applications to external data sources.

Background

Before the Model Context Protocol was established, any AI application that needed to process external data or perform actions beyond the chat interface had to implement a separate interface for each data source. ChatGPT plugins for Google Calendar, Claude tools for Notion, Cursor integrations for GitHub: each combination required its own integration with its own maintenance overhead.

With three AI vendors and ten relevant data sources, this produced thirty specific interfaces. With each additional vendor and each new source, the effort grew proportionally.

What MCP does

MCP is an open standard published by Anthropic in November 2024. It defines a uniform protocol for communication between AI applications and external systems. An AI application implements MCP once, and a data source implements MCP once. The combinatorial complexity is eliminated.

The three roles

An MCP connection involves three parties:

  • MCP Client. The AI application that initiates communication. Examples include Claude Desktop, Claude Web, Cursor and, in time, ChatGPT.
  • MCP Server. The intermediary that connects to an external system. inboxmcp is one such server for IMAP. Other implementations cover GitHub, Linear, Notion and further services.
  • Data source. The actual system being accessed: an IMAP server, a Notion workspace, a calendar.

The client communicates with the server over MCP. The server translates each request into the data source’s native protocol.

What an MCP server provides

  • Tools. Actions the AI can perform, such as sending an email or creating a folder.
  • Resources. Data the AI can retrieve, for example a specific message or the list of available folders.
  • Prompts. Predefined templates for recurring tasks.

Availability on the AI side

  • Claude (Web and Desktop): full support since release.
  • ChatGPT: inclusion in the OpenAI Apps directory has been announced.
  • Gemini: support announced, timeline undefined.
  • Cursor, Continue and other code editors: available today.
  • Microsoft Copilot: on the roadmap.

Significance as an open standard

MCP is not a proprietary product of Anthropic but an open standard with implementations in Python, TypeScript, .NET and Java. Custom servers can be developed and used by any compliant AI application. A binding to a single AI vendor is therefore avoided.

Practical relevance

You do not need to understand the protocol in detail to use it. An AI application is installed, an MCP server is added (enter the URL, complete a one-time OAuth sign-in), and the integration is then available. inboxmcp is one concrete implementation for classic IMAP mailboxes such as mailbox.org, IONOS, STRATO, Posteo and Plesk-hosted mailboxes.

More articles