What is MCP? The “USB-C port” connecting AI to databases, GitHub, Slack…
TL;DR
MCP (Model Context Protocol) is an open protocol that lets AI applications connect to external tools and data in a standard way. An MCP server (e.g. GitHub, Postgres) built once works with any MCP client such as Claude, Claude Code or Cursor.
Three roles
| Component | Role | Examples |
|---|---|---|
| Host | The app users interact with | Claude Desktop, Claude Code, IDEs |
| Client | A 1-to-1 connection to one server, inside the host | Created when you add a server |
| Server | Exposes capabilities to the AI | GitHub, filesystem, Postgres, Slack |
What servers expose
- Tools: actions the AI can call (create an issue, run a query).
- Resources: data to read (files, records, docs).
- Prompts: ready-made prompt templates invoked like commands.
How to put it to work
- Start small: the filesystem or GitHub server is enough to see the value.
- Security: MCP servers run with your permissions — install from trusted sources, use least-privilege tokens, and prefer read-only access to production databases.
- Save context: every server adds its tool list to context; disable the ones you don’t use.
- When building your own: for a single agent workflow a skill may be simpler; MCP shines when connecting systems shared across many apps.