+170 XP

Connectors and the marketplace: plugging Claude into your apps

Ask Claude "summarize my unread email from the last two days and flag anything that needs a reply." With the Gmail connector enabled, it reads your inbox and does it. That is the shift connectors deliver. Claude stops being a smart conversation partner that only knows what you paste in, and starts being an assistant that reaches into your real tools at the moment you ask.

This lesson covers how that works on Claude.ai and what data actually moves. It also shows how to reason about the trust boundary so you can plug Claude into Gmail, Drive, and Calendar without handing over more than you intend.

What a connector actually is

A connector is a pre-built bridge between Claude and an external service. Examples include Gmail, Google Drive, Google Calendar, and a growing list of third-party apps. Claude can call a connector during a conversation when it decides the request needs that data.

Under the hood, connectors are built on MCP (Model Context Protocol), the open standard Anthropic published for exposing tools and data to models. You met MCP at a concept level earlier. The thing to internalize here: a connector is a hosted MCP server that Anthropic (or a vetted partner) runs and lists. You do not have to stand up your own server or write any glue code. You toggle it on, authorize it, and it appears as a capability Claude can use.

The connector directory is the catalog of these connectors inside Claude.ai. You can browse it from Settings → Connectors. Some connectors are first-party (Google Workspace). Many are third-party: project trackers, support tools, knowledge bases. Each one declares what it can read or do before you connect.

The official starting point is the Claude.ai connectors docs. Read the permission summary on any connector before you click Connect.

The concrete flow: "pull the Q3 figure from my Drive doc"

Here is what happens end to end when you ask Claude to fetch something from Google Drive.

You type:

"In my Drive, find the doc called 'Q3 Board Review' and tell me the net revenue retention figure."
  1. Claude recognizes it needs an external tool. Your prompt references your Drive, so Claude proposes calling the Google Drive connector. The first time, Google shows you an OAuth consent screen, the same kind you have approved a hundred times. You grant Claude scoped access.
  1. Claude searches rather than browsing. It sends a search for "Q3 Board Review" to the Drive connector. The connector returns matching file metadata (titles, IDs), and Claude requests the contents of the file that matches.
  1. Only that document's text enters the context window. Claude reads the returned content, finds the net revenue retention number, and answers. Your other thousand Drive files were never opened.
  1. You see the tool call. Claude.ai shows that it used the connector and which document it pulled, so the action is auditable in the transcript.

That last point matters. Connectors are not a background daemon scraping your whole account. They fire on specific requests, fetch specific things, and show their work.

What data is shared, and what is not

This is the part people get wrong, so be precise.

What goes to Claude when a connector runs:

  • The specific items Claude requests for your task. For the email example, that is the unread messages in the window you asked about (subject, sender, body). For Drive, it is the contents of the file or files it opened.
  • The metadata needed to find those items, such as search results, file names, and message IDs.

What does NOT happen:

  • Claude does not get a standing copy of your entire inbox or Drive. It pulls data per request.
  • Connecting Gmail does not mean every future chat silently reads your email. The connector runs only when a prompt calls for it, and you can watch it happen.

Training depends on your plan. On commercial plans (Team, Enterprise, and the API), your content is not used to train Anthropic's models. On consumer plans (Free, Pro, Max), there is a setting that controls whether your chats are used to improve the models. Conversations that pulled in connector data are covered by that setting. Check it under Settings → Privacy. If you opt in, Anthropic also keeps that data longer. Confirm the current terms in the Anthropic privacy docs before connecting a work inbox to a personal account.

The honest caveat: once content enters the context window for a conversation, it is part of that conversation. If a connector pulls an email into a chat and you later share that chat, the email goes with it. Connectors reduce manual copy-paste, but the data they pull still lives in the transcript. Treat a chat that touched real email or financial docs with the same care you would give a forwarded thread.

Scoping access: grant the least you need

When you authorize a Google connector, the OAuth screen lists scopes. Pay attention to read-only versus read-write.

  • A summarize-my-email task needs read access only.
  • A "draft a reply and save it" task needs compose or write access.

Grant the narrower scope when the connector offers a choice. You can revoke any connector's access in two places. One is inside Claude (Settings → Connectors → Disconnect). The other is at the provider, in your Google Account under Security → Third-party access. Revoking at Google cuts Claude off immediately, whatever Claude's own settings show. That makes it the stronger control.

Claude Connectors and Remote MCP Explained

Watch on YouTube

Connectors inside projects and skills

Connectors get more useful when you combine them with the other Claude features you have already learned.

In a Project, you can enable specific connectors so every chat in that Project inherits them. Picture a "Weekly Investor Update" Project with the Drive and Calendar connectors on. You can open a new chat and immediately say "pull this week's metrics doc and my Monday calendar, draft the update." The Project's custom instructions and its connectors give Claude a consistent workspace with the right tools already attached.

With Skills, connectors become the data source for a repeatable procedure. A Skill is a packaged set of instructions (and optionally code) that Claude follows for a recurring task. A "Board Deck Prep" Skill might say: use the Drive connector to fetch the latest financial model, extract the three headline figures, and format them as a summary table. The Skill defines the steps. The connector supplies fresh data each time you run it.

So think of connectors as the input layer. Projects, Skills, and Artifacts decide what to do with the data that connectors bring in.

Knowledge check

1. What fundamental shift does enabling a connector create in how Claude operates?

2. In technical terms, what is a connector 'under the hood'?

3. Why does the lesson recommend reading the permission summary on a connector before clicking Connect?

MULTIPLE CHOICE

4. Select ALL statements that correctly describe the connector marketplace in Claude.ai.

Select all the correct answers.

MULTIPLE CHOICE

5. Select ALL that are true about what happens when you ask Claude to fetch data from Google Drive.

Select all the correct answers.

Custom connectors: when the directory is not enough

The directory covers common SaaS tools, but your company probably runs internal systems too. Connectors are remote MCP servers, so you can add your own.

On Claude.ai you can add a custom connector by pointing Claude at a remote MCP server URL. On paid individual plans (Pro, Max), you add it yourself. On Team and Enterprise, an admin adds it for the organization, and members then authenticate. Once added, your internal tool (a customer database, an inventory API) is available to Claude the same way a directory connector is, and access is gated by your own auth.

A minimal remote MCP server that exposes one tool looks like this:

python
from mcp.server.fastmcp import FastMCP

mcp = FastMCP("inventory")

@mcp.tool()
def get_stock(sku: str) -> dict:
    """Return current stock level for a product SKU."""
    record = warehouse_db.lookup(sku)
    return {"sku": sku, "on_hand": record.quantity, "location": record.bin}

if __name__ == "__main__":
    mcp.run(transport="streamable-http")

Deploy that behind your authentication and register its URL as a custom connector. Now "how many units of SKU-4471 are in stock?" queries your warehouse in real time. The full spec and SDKs for building these servers are at modelcontextprotocol.io and in Anthropic's MCP docs.

Keep two things straight:

  • Directory connectors are reviewed and hosted by Anthropic or a partner. You only authorize them.
  • Custom connectors are your responsibility for security, uptime, and scope. You own that server.

How this differs from the API and Claude Code

You will hear "connector," "MCP," and "tool use" used interchangeably. Sort them by where you use them:

  • Connectors on Claude.ai are the no-code, toggle-in-the-UI version for end users. That is this lesson.
  • The Messages API with tool use is where you, the developer, define tools and run the loop yourself in code. It is a different surface built on the same idea of giving the model capabilities.
  • Claude Code connects to MCP servers from your terminal so Claude can act on your local repo and dev tools.

One mental model carries across all three. MCP is the protocol. Connectors are the click-to-enable version of it inside the Claude apps. What you learn here about scopes and the trust boundary transfers directly to the API and Claude Code, which come next in this path.

A realistic end-to-end task

Put it together. You want a Monday morning briefing.

  1. Create a Project called "Monday Brief."
  2. Enable the Gmail and Calendar connectors on it.
  3. Add a custom instruction: "Every brief should list unread email needing a reply, then today's meetings with any prep needed."
  4. Each Monday, open a chat and say: "Run my Monday brief."

Claude reads the unread mail (read scope), checks today's calendar, and produces the briefing as an Artifact you can edit. There is no manual copy-paste, and the data pulled is visible in the transcript. Nothing left your accounts except what that task required.

Key Takeaways

  • Connectors are hosted MCP servers that you enable from Settings on Claude.ai. They let Claude read live data from Gmail, Drive, Calendar, and third-party apps without any code from you.
  • Data moves per request, not in bulk. Claude pulls only the specific items a task needs and shows the tool call in the transcript.
  • Training depends on your plan. Commercial plans exclude your content from training. On Free, Pro, and Max, check the training setting under Settings → Privacy.
  • Scope tightly and revoke at the source. Grant read-only when that is all the task needs. Revoking in your Google Account security settings cuts access immediately and completely.
  • Combine connectors with Projects and Skills to build repeatable workflows like a Monday brief.
  • When the directory falls short, add a custom connector as a remote MCP server. This works on Pro and Max as well as Team and Enterprise, but you own its security and scope. The same MCP model carries forward to the API and Claude Code.

Frequently asked questions

What is a Claude connector?

A Claude connector is a pre-built bridge between Claude and an external service like Gmail, Google Drive, or Google Calendar. Technically it is a hosted MCP server (Model Context Protocol) run by Anthropic or a vetted partner, so you enable it from Settings → Connectors and authorize it without writing any code. Claude then calls it during a conversation when your request needs that data.

Does connecting Gmail to Claude give it access to my whole inbox?

No. Claude pulls data per request, not in bulk: if you ask for unread email from the last two days, it fetches those messages and nothing else. The connector only runs when a prompt calls for it, and Claude.ai displays the tool call in the transcript so you can see exactly what was retrieved.

Are conversations that use Claude connectors used to train Anthropic's models?

It depends on your plan. On commercial plans (Team, Enterprise, API), your content is excluded from model training. On Free, Pro, and Max, a setting under Settings → Privacy controls whether your chats are used to improve the models, and conversations containing connector data fall under that same setting; opting in also means Anthropic retains the data longer.

How do I revoke a Claude connector's access to my Google account?

Two places work, but they are not equivalent. You can disconnect inside Claude at Settings → Connectors → Disconnect, or revoke at the provider in your Google Account under Security → Third-party access. Revoking at Google is the stronger control because it cuts Claude off immediately, whatever Claude's own settings show.

What is the difference between connectors on Claude.ai, tool use in the API, and Claude Code?

They are three surfaces built on the same idea of giving the model capabilities. Connectors on Claude.ai are the no-code version you toggle on in the interface. The Messages API with tool use is where a developer defines the tools and runs the loop in code. Claude Code connects to MCP servers from your terminal so Claude can act on your local repo and dev tools. MCP is the underlying protocol in all three cases.

How do I connect Claude to an internal company system that is not in the connector directory?

You add a custom connector by pointing Claude at the URL of your own remote MCP server. On Pro and Max you add it yourself; on Team and Enterprise an admin registers it for the organization and members then authenticate. Unlike directory connectors, which Anthropic or a partner reviews and hosts, you own the security, uptime, and scope of a custom connector.

What to do, from this lesson

These actions are compiled in the role's Playbook.

  • Grant read-only connector scopes and revoke tokens at the provider quarterly
  • Name the tool sequence and sources explicitly when chaining connectors
See the full action playbook →

Related articles

Recent articles from the blog that build on this lesson.