MCP: The Connective Tissue
Connectors are what let Claude read your Gmail, pull a doc or query QuickBooks — in plain English, no developer. Here's what MCP is, and the one question to ask before you plug anything in.
"USB-C for AI" — one standard plug
The Model Context Protocol is an open standard that gives AI tools a common way to exchange context and actions — so the "many apps × many tools" mess collapses to "many + many". Toggle the view.
For a small business this is the unlock: you don't integrate anything. You click "connect", approve it, and ask in plain English. Anthropic's directory spans hundreds of connectors — and the number keeps moving, so never pin a course or a plan to a fixed count.
Connector Capability Explorer
Capabilities vary — that's the whole safety story. Some connectors only read, some can draft, some can act with your approval. Filter by what you'd let them do.
Capabilities are indicative as at June 2026 and differ by connector version and region — always check what a connector can do before enabling it. Examples: the Gmail connector drafts but typically won't send for you; Google Workspace actions need per-action approval.
The one question: read, draft, or act?
Before you enable any connector, know which of these it does. It's the foundation of the "reads before writes" build you'll do in Module 3.
It looks, never changes
Pulls reports, searches your inbox, reads a doc. Pure upside — no risk of breaking anything. Start every hire here.
It prepares, you send
Writes the reply, builds the invoice, assembles the summary — and waits. You review and hit send.
It does — with approval
Books the meeting, posts the message. Each consequential action needs your explicit OK. Never unattended on money or sensitive data.
The rule of the course: automate the reads, supervise the writes. Read-only is where every safe AI hire begins.