Connectors

Claude Cowork Connectors: The Complete Guide

Connectors link Claude Cowork to Gmail, Slack, Notion, Google Drive, and more. Here's how they work, how to set them up safely, and how to use them in real workflows.

Connectors are the single feature that makes Claude Cowork a genuinely new category of tool rather than a better chatbot. They are also the feature most people set up wrong on the first try — too permissive, too many, or too few. This guide covers what connectors are, how to install them, how to use them in real work, and how to set permissions so you don’t have a bad day.

What a connector actually is

For most of the past few years, AI chat had two problems: a first-mile problem (getting your data in) and a last-mile problem (getting its output back into your real tools). You solved both by copy-pasting. The AI was smart; the workflow was awful.

A connector fixes both. It is a pre-authenticated, permissioned link between Claude and another piece of software. Once you’ve connected Gmail, Claude can search your email. Once you’ve connected Notion, Claude can create pages. Once you’ve connected Granola, Claude can pull meeting transcripts.

Under the hood, connectors are built on a technology called MCP (Model Context Protocol). MCP takes natural language and converts it into structured API calls that the other app understands. You do not need to know how MCP works. You do need to know that connectors exist, what they can do, and how to permission them.

The shape of the connector catalog

Connectors broadly fall into four categories:

Communications

  • Gmail, Outlook, Microsoft 365
  • Slack
  • Granola (meeting transcripts)
  • Fireflies (meeting notes)

Productivity and knowledge

  • Google Drive, Google Calendar
  • Notion, Atlassian (Confluence, Jira)
  • Airtable, Smartsheet
  • Asana, Monday, ClickUp
  • GitHub

Business systems

  • HubSpot, Salesforce (and other CRMs — slower to arrive, but coming)
  • Stripe
  • Ramp
  • DocuSign
  • Gamma, Canva, Figma (for creation)

Industry-specific

  • S&P, FactSet, Capital IQ, PitchBook (finance)
  • Harvey (legal)
  • And a growing list of vertical tools

The catalog grows weekly. The list above is a snapshot of what’s available today and will be incomplete by next quarter.

Installing a connector

  1. Open Cowork and click the Customize button (it looks like a small briefcase or gear icon).
  2. Click Connectors.
  3. You’ll see two sections: Web connectors (Gmail, Slack, etc.) and Desktop connectors (more technical, usually tied to Claude Code workflows).
  4. Click the + icon, then Browse connectors.
  5. Scroll or search the catalog. When you find one, click Connect.
  6. You’ll be redirected to the source app (Google, Slack, etc.) to authorize the connection.
  7. Grant only what you actually need.

One note on permissions at this stage: connecting Gmail does not give Claude access to your entire company’s email. It gets access only to what your account can see — the same scope your own login has. If you’re on a team plan, your administrator may have locked down which connectors you can install at all.

Turning connectors on for a session

Connecting an app is different from using it in a conversation. To make a connector available in a specific Cowork session:

  1. Click the + icon inside the chat.
  2. Click Connectors.
  3. Toggle the ones you want active on.

You’ll see a Tool access setting: “Load tools when needed” vs “Always load tools.” Start with Load tools when needed — this keeps your context window lean. Tools pulled in on demand use fewer tokens overall.

Using connectors in plain English

You do not need special syntax. Ask for what you want. Some starter examples:

What channels are in my Slack group?
Go through my email inbox from the past 24 hours
and surface any important messages that need a reply.
Pull up my last five Granola meetings and write a
summary of open action items.

One real lesson from the above: be explicit about which inbox you mean. “Check my inbox” is ambiguous — you have a Slack inbox, an email inbox, and a Notion inbox. Say “email inbox from Gmail” and you’ll save a round trip.

Combining connectors

The real power shows up when you combine connectors in a single natural-language request:

Pull up my last five meetings in Granola. Create a new
Notion database and add the following fields: date,
topic, meeting type, action items. Populate it with
the five meetings.

Cowork will: call the Granola tool to list meetings, call the Granola tool again to fetch each transcript, tag the meetings by type, create a Notion database with the right schema, then write five rows into it. You’ll see each step in the chat with a permission prompt for new actions you haven’t authorized yet.

This is the “chief of staff” pattern. Combine communication tools + knowledge tools + task tools and you have something that genuinely looks like an assistant, not a chatbot.

Under the hood: reads vs writes

Every connector exposes a set of tools. For Gmail, the tools are:

Reading tools

  • Get profile
  • Get list of drafts
  • Get list of labels
  • Read Gmail thread
  • Read Gmail message
  • Search Gmail (the most used)

Writing tools

  • Create Gmail draft (the only write action — Claude cannot send)

For Google Calendar, the write side is larger:

Reading tools

  • Find meeting times, find free time
  • Get event details
  • List calendars, list events

Writing tools

  • Create event
  • Delete event
  • Respond to event
  • Update event

The pattern holds across the catalog: there are always more read tools than write tools, and write tools are intentionally narrow. Claude can draft emails but not send. Claude can create Notion pages but not delete databases. This is by design.

Setting permissions the right way

For every tool on every connector, you (or your administrator) can set one of three permission levels:

  1. Always allow — Claude uses this tool without asking.
  2. Needs approval — Claude shows you the action and waits for your click.
  3. Blocked — Claude cannot use this tool at all.

A sane starting discipline:

  1. Block all write tools by default. If a connector doesn’t need to create or edit anything for your workflow, block the whole write side.
  2. Set all read tools to Needs Approval. You’ll see exactly what Claude is reading before it reads it. Tedious for the first week, fast once you see the patterns.
  3. Graduate specific tools to Always Allow as you get comfortable. “Search Gmail” is a good first Always Allow. “Create Gmail draft” should probably stay at Needs Approval forever.

If a tool makes you nervous, keep it at Needs Approval. The friction cost is five seconds per call. The cost of a wrong confident action is much higher.

Common mistakes

  • Installing every connector on day one. You’ll spend the afternoon permissioning things you don’t use. Install one at a time, use it for a week, then add the next.
  • Leaving all writes on Always Allow. This is the setup most tutorials show. It is not the right default. Claude is good, but “agent with write access” deserves more friction, not less.
  • Being vague about which tool you want. “Check my inbox” is ambiguous when you have Slack, email, and Notion connectors all active. Name the source.
  • Forgetting connectors are off by default in new chats. A fresh Cowork chat may not have your connectors toggled on. Check the + menu before asking a connector-dependent question.

Putting it together: a connector-heavy session

Here is a realistic end-to-end prompt that uses three connectors:

1. Check my Gmail for anything flagged urgent in the
   last 48 hours.
2. Cross-reference the senders against my Notion CRM
   and tell me which are existing clients vs prospects.
3. For each existing client, check my Google Calendar
   and surface any upcoming meetings with them.
4. Draft a reply to each existing client acknowledging
   their email and referencing our next meeting if any.

Cowork will chain the three connectors, pause for your approval where needed, and hand you back a list of drafts and a context summary. You review, edit, click send. That is what connectors unlock.

Next up

Once you have two or three connectors working, the next concept to learn is projects — self-contained workspaces with their own files, memory, and instructions. Connectors bring in the live data; projects hold the reference material and the rules of engagement.

Frequently asked questions

What is a connector?

A connector is a pre-built bridge between Claude and another app — Gmail, Slack, Notion, Google Drive, and so on. Once connected, you can ask Claude in plain English to read from or write to that app. Under the hood it's a technology called MCP (Model Context Protocol), but you don't need to think about that to use it.

Is it safe to give Claude access to my email?

Yes, when configured correctly. For every connector, each individual action (read, search, create, delete) has its own permission level: Always Allow, Needs Approval, or Blocked. The safest default is to block all write actions, set all read actions to Needs Approval, and loosen specific ones as you build trust.

Can Claude actually send an email for me?

No. For Gmail, the only write tool available is Create Draft. Claude cannot click send. You always review the draft in your own inbox before it goes out.

What connectors are available?

The list keeps growing. Common ones: Gmail, Google Drive, Google Calendar, Slack, Notion, GitHub, Granola, Asana, Monday, HubSpot, Stripe, Figma, Airtable, PitchBook, Canva, WordPress, DocuSign. Industry-specific ones (S&P, FactSet, Capital IQ, Harvey for legal) are expanding quickly.

Do connectors work differently in Chat vs Cowork?

The same connector catalog is available in both, but Cowork can combine connector output with files on your local machine. That's the crucial difference: a Cowork session can pull five Granola transcripts, then write them into a Word doc that lives in your Documents folder.

Get the free Cowork Insider playbook

The 20-minute step-by-step guide for non-technical professionals. Delivered by email.

No spam. Unsubscribe anytime.