This lesson is being recorded
Every video of this course is being uploaded to my YouTube channel. Subscribe so you do not miss it.
Bruno Ashford @brunoashfordSubscribeModule 6 · Lesson 6.1 · 11 min
Connectors and MCP, Plain English
Understand how an agent reaches your email, calendar, drive and CRM, and what MCP actually is.
Use ← → to move. The marker highlights text and the pencil draws on the slide; navigation locks while you annotate. Undo removes the last mark, Esc exits.
1Think of MCP as a set of labelled buttons
Each MCP server hands Claude a list of tools, like buttons on a panel: search contacts, read a document, create a draft. Claude can only press the buttons the server offers, and you decide which buttons need your approval. Nothing magical happens outside that panel.
2Know the two ways a connection arrives
The easiest path for owners is a connector you add in your claude.ai account settings. When you are logged in to Claude Code with the same account, those connectors show up automatically. The second path is adding a server directly in the terminal with the claude mcp add command, usually with a URL the vendor publishes.
3See what is already connected
Before adding anything, look at what you have. The list command shows every server and whether it is healthy. Inside a session, the /mcp command opens a panel where you can check status, sign in and see each server's tools.
claude mcp list4Understand scope: who gets the connection
A connection can be local (just you, just this project), project (shared with the team through a .mcp.json file in the project folder) or user (you, across all your projects). For anything that touches client data, start local. Share with the team only after the connector has passed your risk check.
5Ask Claude to explain its own panel
Once a server is connected, ask Claude to list the tools it can now use and to flag which ones change data. This is the fastest way to learn what you just plugged in, and it tells you exactly where approvals matter.
List every MCP tool you can use right now. Group them by server, and for each one tell me in plain English whether it only reads data or whether it can create, change, send or delete something.A roofing contractor I worked with had Claude write beautiful follow-up emails for two weeks. Then he realised he was still copying lead names out of his CRM, pasting them into the chat, and pasting the drafts back into Gmail. The AI was fast. He was the slow part in the middle, acting as a human USB cable.
Connectors remove that human cable. MCP, the Model Context Protocol, is simply a standard plug. A tool like HubSpot, Notion or your Google account publishes an MCP server, and Claude Code plugs into it. Once connected, the agent can look things up and take actions inside that tool, within the permissions you allow.
Key points
- MCP is a standard plug; each server gives Claude a fixed panel of tools.
- Owners usually connect through claude.ai connectors; the terminal route uses claude mcp add.
- Scope decides who shares a connection: local, project or user.
- Always separate read tools from tools that change or send.
Avoid
- Connecting five tools on day one before you know what any of them can do.
- Sharing a connection with the whole team at project scope before testing it alone.
- Assuming a connector only reads because you only asked it to read.
A three-chair dental clinic connected its Google Calendar and Gmail through claude.ai connectors. On the first day the practice manager only asked Claude to list the tools and mark which ones could send or change anything. That five-minute exercise produced the clinic's first rule: Claude may read the calendar and draft reminder emails, but a human presses send. Two weeks later the manager was saving about 40 minutes a day on reminder and rebooking emails, without a single message going out unreviewed.