Two words first
- Connector (MCP): Claude plugs into an app you already use, like Gmail, Google Drive or Notion. It can read from it and, if you allow it, change things there. MCP is the open standard behind connectors. Anthropic compares it to a USB-C port for AI apps.
- Skill: a folder of instructions (a SKILL.md file, sometimes with scripts) that teaches Claude how to do one job well. Anthropic puts it this way: connectors give Claude access to tools, skills teach Claude how to use them.
Don't start from someone's "top 5" list. Most of those are tools for programmers. Start from your own week.
Prompt 1: Research (what do I actually need?)
Swap the words in [brackets] for your own, then paste:
I want to use Claude for [my work, e.g. answering client emails and planning my week]. Research which connectors, MCP servers or skills would actually help. Give me 3 options at most. For each: what it does in one line, who makes it, what it can read and what it can change, free or paid, how to add it. Only suggest official ones or ones made by the app's own company. Then tell me which ONE to start with, and why.
You get a short table. Pick the one it recommends. Leave the other two for later.
Prompt 2: Add (only that one, safely)
Walk me through adding [tool] in the Claude app, step by step. Tell me which permissions it will ask for, which ones I can say no to, which actions should always ask me first, and how to disconnect it later.
In the Claude app, connectors live in Customize › Connectors. Pick the one, press Connect and sign in to that app. Skills live in Customize › Skills (Code execution has to be on in Settings › Capabilities).
Prompt 3: Learn (use it well, today)
I just connected [tool]. Teach me to use it well for [my work]: 5 things I can ask it today, the mistakes beginners make, and what it should never do without asking me. Then make it a one-page cheat sheet with copy-paste prompts.
Save the cheat sheet in a note. Try two of the five asks today.
The safety check (30 seconds, before you press Connect)
- Who makes it? Official connectors from the directory (claude.ai/connectors) or from the app's own company. Skills only from sources you trust. Anthropic's own advice: install skills only from trusted sources, and read the files before you turn one on.
- What can it read, and what can it change? Reading your calendar is one thing. Sending email in your name is another. Claude gets the same access you have in that app.
- Keep risky actions behind your approval. Anything that sends, deletes or pays should wait for your yes. Claude's own Gmail connector already works this way: it can send, reply to and forward emails, and by default it asks for your approval before each of these. Same for sharing, moving or trashing files in Google Drive. When Claude asks, read what it is about to do before you approve it. Don't switch off the asking for tools that send or delete. You can also turn off tools you don't need: press + in the chat, hover over Connectors and toggle them.
The honest part
Menus and plan limits change. This is how it looks as of October 2026 (on the Free plan you can add only 1 custom connector). Connectors and skills can be tricked by text they read (prompt injection), so start with one tool, keep approvals on, and add the next one only when the first one is useful.
Sources: modelcontextprotocol.io (intro), support.claude.com articles 11176164, 10166901, 11175166, 12512176, 12512180.