Claude Connectors: How to Connect Your Apps to Claude

By default, Claude knows only what you train it on or type into the conversation. That works for a lot of tasks. It does not work when you need Claude to check what actually came in overnight on Slack, find the latest version of a file in Google Drive, or draft a reply to a specific Gmail thread. For that, you need a connector.

Claude connectors link Claude to your external apps and services so it can retrieve live data and take actions inside those services during a session. They are different from uploading a file, which is a one-time static snapshot. A connector gives Claude an ongoing, authenticated channel to an app, with the ability to read or act in real time.

What a connector actually does

A connector does two things. First, it retrieves data from an app you have authorized. Second, depending on the connector, it lets Claude act inside that app on your behalf.

The permission model is worth understanding before you add anything. Claude inherits your permissions from the connected service. If you can see a file in Google Drive, Claude can see it when the connector is active. If a folder is restricted so your account cannot open it, Claude cannot open it either. Claude operates within your existing access, not above it. You are not granting Claude new power in the service; you are letting Claude act as you within the limits you already have.

You authorize each connector separately. That means you go through an OAuth flow or equivalent authentication for each service you want to connect. Adding a Slack connector does not automatically give Claude access to Gmail. Each one is a discrete authorization you control and can revoke at any time.

How it differs from uploading a file

Uploading a file to a Claude conversation gives Claude a static copy of that document at that moment. Claude reads what you pasted or attached. That is it.

A connector is different because the data stays live. If you ask Claude to search Google Drive through a connector, it queries Google Drive at that moment and returns current results. If a document was updated an hour ago, Claude sees the updated version. If new emails arrived since you started the session, a Gmail connector can pull them. The difference matters whenever you are working with things that change.

It also matters for actions. You cannot “upload a file” and have Claude reply to an email. With a Gmail connector, Claude can draft or send a reply inside Gmail. The connector is a two-way channel, not a one-time import.

Where to find and add connectors

The directory of available connectors is at claude.ai/connectors. Current named examples include Slack, Google Drive, Gmail, Linear, Microsoft 365, and DocuSign. The list is growing, so the directory is the most current reference.

Adding a connector takes a few steps. Go to the connectors directory, find the service you want, click to connect it, and authenticate with that service. After that, the connector is available across Claude, Claude Desktop, Claude Cowork, and the API. You do not need to re-add it for each context.

On the free plan you can add one custom connector. Paid plans (Pro, Max, Team, Enterprise) support more. For most people building a real workflow, a paid plan is the practical path once you have more than one service to connect.

What you can do with them in practice

The clearest way to understand connectors is through concrete tasks, not abstractions.

With a Slack connector, you can ask Claude to read a specific thread, summarize activity across channels, or draft a message to post. Without it, you copy and paste.

With Google Drive, Claude can find a document, read its current contents, and work from it directly. Without it, you download the file and upload it to the conversation.

With Gmail, Claude can read threads, draft replies, or search your inbox for messages matching specific criteria. With Linear, it can read tickets, update statuses, or summarize what is in progress. DocuSign lets Claude interact with document workflows.

The pattern in each case is the same: Claude goes and gets the actual thing, rather than working from a copy you provided.

In Claude Cowork specifically, connectors are central to how tasks work. Cowork is built around agents completing multi-step work, and those agents need access to your tools to do it. A Cowork task might read a Slack thread, search Google Drive for supporting documents, and draft a Gmail reply, all in one session, with each of those steps going through the relevant connector. The connector is what makes the agent useful rather than generic.

For more on how connectors fit into Cowork alongside skills and scheduled tasks, see Connectors, Skills, and Scheduled Tasks.

The MCP connection

Claude connectors are built on MCP (Model Context Protocol), the open standard for connecting AI models to external tools and data. Anthropic published MCP as an open specification, which means it is not proprietary to Claude connectors specifically. Any model that implements MCP can work with MCP-compatible tools. Any tool built to the MCP spec can work with any compatible model.

For most users this is background information. You do not need to understand MCP to add a Slack connector and use it. But it matters if you are building custom integrations or thinking about the long-term portability of what you set up. A connector built to an open standard is not locked to a single vendor’s decisions. You can read more about the underlying standard at What Is MCP.

A note on safety and scope

Connectors are genuinely useful. They are also worth adding thoughtfully.

The practical principle: only connect what a current task actually needs. If you are drafting a client proposal and need Google Drive for reference documents, add that connector. If you are not actively using Gmail in a session, there is no reason to have it active. Fewer active connections mean a smaller surface for anything unexpected to touch.

It is also worth being specific with Claude about what you want it to do in a connected app. “Search my Drive for the Q2 report” is a well-scoped request. “Look through everything in my Drive and find anything relevant” is broad in a way that may produce more than you intended.

Claude operates within your permissions, which is a real safeguard. But the most useful safeguard is being clear about what you want done and checking the output before anything is sent or changed.

Getting started

If you have not added a connector yet, the directory at claude.ai/connectors is the starting point. Pick one service you already use and that you want Claude to have access to during a task. Add it, test it with a small request, and see what the workflow looks like before adding more.

If you are using Cowork and want to see connectors in the context of a full agent workflow, the Cowork beginner course walks through setting up a connected task from scratch.

For a broader look at what plugins and skills add on top of connectors, see Claude Cowork Plugins and Claude Cowork Skills.