Summary  
This chapter demonstrates how to integrate external service connectors into an application by implementing OAuth-based authentication flows, managing permission scopes, and configuring connector-specific settings.  

General domain of usage  
Email management

Now it's time to plug things in. You'll connect Gmail because it's one of the most universal connectors, and the same flow works for other Google apps like Calendar, Drive, Docs, and Slides, as well as most first-party connectors in Claude's catalog.

Learn the process once, and you can reuse it almost everywhere.

Before adding a connector, get oriented inside Cowork so you know where everything lives.


The three top-level modes. This course focuses on Cowork, where you build actual agents.

Each task is its own AI agent. Use a separate task for each specialized workflow to keep agents focused and easier to debug.

Folders that group related tasks together. Use them when you have multiple agents working in the same area.

Lets you run an agent daily, weekly, or on another schedule without setting up a cron job. This makes the agent more autonomous.

A place for generated HTML files that you want to host publicly. You will not use this much in the course.

Turns your computer into a remote agent server. If your computer is on, Cowork can execute tasks from your phone while you are away.

The place where you connect apps, create skills, and browse plugins. This is where you are headed next.

When you open a new task in Cowork, there are a few settings worth getting right up front:

Click "Choose folder" and pick (or create) a folder on your computer where this agent can output files. Mine is called "AI Agent Course." Any markdown summaries, generated docs, or files the agent produces will land here. Gives the agent a workspace.

Working Folder

Keep this on "Ask before acting." This is the most important safety setting in the whole app for your first month with Cowork. With it on, the agent will pause and ask you before doing anything that changes data — sending a message, deleting a file, modifying a doc. Once you've built up trust and understand exactly what your agent does, you can switch to "Act without asking." But not yet. Treat the ask-before-acting toggle like training wheels you don't take off for a while.

Permission Mode

Sonnet 4.6 is the right default. It's fast, capable, and easy on your usage limits. Opus 4.7 is more powerful but burns tokens fast. Save Opus for the harder workflows later.

Model

### Adding the Connector

Click the **plus icon** in the bottom-left corner of the chat screen.

Open **Connectors → Manage Connectors**. This is your dashboard for everything you have connected.

Click the **plus icon** at the top of that screen, then select **Browse Connectors**. This opens the connector marketplace.

Choose **Gmail**. A Google authentication window will open in your browser.

Sign in to your Google account and approve the permissions. Once authentication is complete, you will return to Cowork, and Gmail will be connected.

### The Permission Panel

After you add a connector, Claude shows a detailed permission panel. This panel controls exactly what Claude can do inside that app.

For Gmail, permissions are usually split into three tiers:

Claude can read emails without asking every time. This is safe because reading does not change anything.

Claude must ask before drafting emails, modifying labels, archiving threads, or deleting anything.

Claude cannot send emails automatically. You can still let it draft emails, then review and send them yourself.

Every connector has its own permission panel. Review each one carefully before using the connector.

This permission system is what keeps your agents inside the lines. Five minutes of setup now can prevent mistakes later.

Note

What is the safest recommended setting for Gmail actions like drafting, archiving, or deleting emails?

Learn how to use AI agents that do more than answer questions. This course shows you how to set up Claude Cowork, Claude for Chrome, connectors, scheduled tasks, and mobile control through Dispatch. You will learn how agents interact with apps, browse the web, organize files, create documents, build spreadsheets, and complete multi-step workflows. By the end, you will understand how to design safe, useful agents that fit your own work instead of just following demos.

Learn what AI agents are, how they differ from chatbots, and where they are most useful.

Explore the main agent tools available today and decide which setup fits your needs.

Connect Claude to the apps you already use and learn how to manage permissions safely.

Use agents to complete real multi-step tasks across Drive, Docs, Sheets, Chrome, and Shopify.

Take your agents beyond the desktop, design a useful keeper agent, and understand where agent tools are heading.

Connecting First-Party Apps

Adding the Connector

The Permission Panel