MCP and connectors: plugging in other apps
Module 9 — MCP and connectors: plugging in other apps
← Module 8 | Index | Next: Module 10 →
Out of the box, Claude can work with files and run commands. But what about your email, your calendar, your project tracker, your company database? That’s what MCP is for.
MCP (Model Context Protocol) is a standard way to connect Claude to outside tools and data. A connector (or “MCP server”) is one such connection — to Slack, Gmail, Google Drive, GitHub, a database, and so on. When a connector is attached, Claude gains new tools: it can now read your Slack messages, create calendar events, query your database — whatever that connector exposes.
Why it’s a big deal
It turns Claude from “an assistant for my files” into “an assistant for my whole working life.” Instead of you copy-pasting a Slack thread into the chat, Claude reads it directly. Instead of describing a ticket, Claude pulls it from your tracker.
How you use them in practice
- In Cowork, when you ask for something that needs an outside app (“what’s on my plate this sprint?”), Claude searches a registry of available connectors and suggests the right one to connect. You approve it, and it’s wired in.
- In Claude Code, you add MCP servers through configuration; they then appear as extra tools.
- Connectors come with permissions — you control what Claude may read or change. Treat write access (sending messages, moving money, deleting things) with care.
A safety note worth internalizing
Content that comes in from the outside world — an email, a web page, a message — should be treated as untrusted. A well-designed setup never lets Claude blindly act on instructions hidden inside fetched content. Be especially cautious approving connectors that can send or delete, and never let automated steps move money or make trades on your behalf.
The mental model: each connector is a new doorway. Claude can only walk through the doors you open, and only do what that door allows.