Connecting to External Services with MCP (Model Context Protocol)

What Is MCP?

MCP (Model Context Protocol) is an open protocol specification for connecting AI models to external tools and data sources. Proposed by Anthropic, it has been adopted by other AI providers and the broader developer community as well source.

With MCP, you can describe integrations — such as "call this operation on this service from Claude" — using a unified protocol. There is no need to write custom integration code for each individual service. Once you implement an MCP server, it can be used from any MCP-compatible client, including Claude Code, Claude Cowork, and the API.

Basic Structure of MCP

MCP consists of three layers: "host," "client," and "server."

• MCP Host: An application — such as Claude Code, Claude Cowork, or a custom app — that contains an MCP client.
• MCP Client: The component within the host responsible for communicating with MCP servers.
• MCP Server: A server that wraps the APIs of external services — such as GitHub, Zoom, Google Drive, or Slack — using the MCP protocol. Custom implementations are also possible.

Examples of Common MCP Connectors

A wide variety of MCP servers have been published by official sources and the community. Notable examples include:

• GitHub MCP Server — Manage repositories, issues, and pull requests
• Google Drive / Docs — Browse files and generate documents
• Zoom — Retrieve meeting information and manage schedules
• Slack — Send messages and browse channel information
• Database connections — Run queries against PostgreSQL, SQLite, and more

For details on available connectors, refer to the official documentation (modelcontextprotocol.io).

Using MCP in Claude Cowork

Claude Cowork achieves integration with external services through MCP connectors. For example, tasks like "summarize the Zoom meeting minutes" or "list GitHub issues and prioritize them" are handled by Cowork accessing external data via MCP.

Combining with Plugins

Plugins, covered in the previous lesson 3-4, can include MCP servers as one of their components. This makes it possible to "distribute connection settings to external services along with the plugin to your team," saving each member the effort of individually configuring MCP servers.

Level 3 Complete

Congratulations. You have completed all 5 lessons in Level 3: "Advanced Features." You now have a comprehensive understanding of the advanced features that help you get more out of Claude — Artifacts, Computer Use, Skills, Plugins, and MCP. Level 4, "Claude Code," covers the introduction and practical use of Claude Code, the CLI tool for developers.

MCP Dedicated Article (Coming Soon)

Implementation methods for MCP, how to build your own MCP server, and security considerations will be covered in detail in the MCP dedicated article (coming soon) on Clauder Navi. A link will be added to this lesson once it is published.