Connectors — Giving Co-Work Access to Your World

What Connectors Are (and Are Not)

Connectors are MCP-based integrations that give Co-Work direct access to your tools. No code required. You authenticate through an OAuth flow — the same "log in with Google" pattern you have used hundreds of times — and Co-Work gains the ability to read from and write to that application. Approximately 50 to 70 native connectors are available at launch, covering everything from Google Workspace and Notion to Salesforce and Slack.

Understanding what connectors are not is just as important. Connectors are not skills (reusable instruction sets), not plugins (bundled capabilities), and not the same as third-party automation tools like Zapier. They are the raw integration layer — the plumbing that lets Co-Work reach into your actual tools and operate them.

The Permission Trust Model

Connecting an app to Co-Work is not a single binary decision. Every connector involves a permission level — read, write, or execute — and those three levels carry fundamentally different risk profiles. Think of it the same way you think about giving a new employee access to your tools.

Level	What It Allows	Risk	Guardrail Required
Read	Co-Work can see data — emails, events, files — but cannot change anything	Low	None beyond standard global instructions
Write	Co-Work can create and modify — draft emails, create documents, add calendar events	Medium	Review-before-send rule in global instructions
Execute	Co-Work can take actions — send emails, delete files, post to Slack	High	Explicit approval required for every irreversible action

The practical rule: start everything at read permission. Test that Co-Work can access the data you expect. Only then upgrade to write, and only after your global safety guardrails are in place. Execute permission — the ability to send things and take irreversible actions — requires your safety guardrails to be rock solid.

Connector Data Privacy

This section comes directly from official Co-Work documentation, and it is often omitted from community content. Read it carefully before connecting any app that handles sensitive data.

Data you send through connectors flows through Claude's systems. When you ask Co-Work to check your Gmail and summarize your inbox, your email content passes through Anthropic's systems as part of the request. The same applies to calendar data, documents, and anything else a connector surfaces.

The practical implication: do not connect applications that contain regulated data — HIPAA-covered health information, PCI-DSS payment card data, financial records subject to SOX — until you have read Module 16 and understand the compliance boundary. Co-Work is not certified for these use cases. Module 16 covers what Co-Work can and cannot do in regulated contexts in full.

The Five Essential Connectors for Most Learners

You do not need every available connector. For most knowledge workers and small business operators, five connectors cover the majority of automation use cases in Modules 5 through 12:

1. Gmail — email triage, draft creation, thread monitoring, inbox summarization. Start with read + draft only.
2. Google Calendar — event reading, scheduling analysis, conflict detection. Read permission is sufficient for most tasks.
3. Google Drive — document reading, file creation, output storage. Required for saving task output to Drive folders.
4. Notion (if you use it) — note creation, database updates, project page management. Use for output storage from research and content tasks.
5. Slack (if you use it) — channel monitoring, draft message creation. Never grant post-without-confirmation permission in early use.

Connect one at a time. Test each with a read-only task before building any automation on top of it. A connector that appears connected may not have the correct OAuth scope — the only way to know is to test.

Chrome as the API-Less Alternative

Some apps you need do not have native connectors and do not expose clean APIs. For those cases, Co-Work can control Chrome browser — navigating to websites, reading content, filling in forms — the same way a human would. Paul calls this the Chrome alternative; it is documented in official Co-Work capabilities.

When to use Chrome instead of a native connector:

• The app you need is not in the native connector library
• The task involves reading public web data (competitor prices, news, job postings)
• The app has no MCP server available and Zapier does not integrate with it

Chrome automation is slower than API connectors and fragile when a site's UI changes. Use it for low-frequency tasks — competitive research, occasional social monitoring, one-off web lookups — not high-frequency automated tasks that run daily. For those, find an API-based solution first.

Zapier MCP: The Long Tail

If your app is not in the native connector library and Chrome is too fragile for your use case, Zapier MCP extends Co-Work's reach to over 8,000 applications. The setup involves creating a Zapier MCP server, selecting the tools you need, and pasting the server URL into Co-Work's connector settings. Jack Roberts covers this process in detail in the Skills course.

Two things to factor in before choosing Zapier MCP: it requires a Zapier subscription (factor into your automation ROI), and the data privacy note from above applies here as well — data flowing through Zapier also flows through Zapier's systems, not just Anthropic's. For apps handling sensitive data, consult Module 16 before proceeding.