Claude Connectors: What They Can Read, Change and Reach

A connector turns Claude from something you paste information into into something that can go and get it. Instead of copying a document out of your drive, you let Claude search the drive. Instead of describing a ticket, it opens the ticket. That convenience comes from the fact that Claude is now reading your accounts, and in some cases changing them.

This article explains what Anthropic documents about connectors, what the permission levels mean, and a short checklist to run through before you connect anything. For the wider picture of Claude, the Complete Beginner's Guide to Claude is the place to start.

What a connector is

Anthropic describes connectors as a way for Claude to access your apps and services, retrieve your data and take actions in them. Claude uses each person's existing permissions in the connected service, so it sees what you could see and nothing more. Connectors work in Claude, Claude Desktop, Claude Code and the API through the MCP Connector. Anthropic's examples include Linear, where Claude can create issues, Slack, where it can send messages, and Google Drive, where it can search files.

"MCP" stands for Model Context Protocol, the open standard these connectors are built on. You do not need to understand it to use one, but you will see the term in settings.

Reading is not acting

The most important line in any connector's listing is whether it can read, write or both. Anthropic's directory shows this for each connector.

Start with read-only wherever you have the choice. Add write access only for a specific job where you have thought about what could go wrong.

Adding a connector

The steps are short. Open the Connectors Directory from a chat, using the "+" or "/" menu, or from Customize and then Connectors. Select a connector, read its description, click Connect or Install and complete the authentication. Some connectors need extra settings or permissions during setup.

Once connected, you switch a connector on per conversation with the toggle in the same menu. Claude can also bring a connected app into a conversation on its own when it fits the request. Some connectors, marked "Interactive," can show live interfaces such as dashboards or task boards inside the chat.

Anthropic also describes two tool-access modes. Auto is the default, and "on demand" is suggested when ten or more connectors are active. Past that point, switching to on demand is Anthropic's suggestion.

Team and Enterprise controls

On Team and Enterprise plans, an Owner or Primary Owner has to enable a connector for the organization first, and enabling it does not give anyone access: each user still authenticates individually. Owners can set tool permissions by category or by individual tool: always allow, needs approval or blocked. Those restrictions apply across the organization and individual users cannot override them, and they can only narrow access, never grant more than the source system allows.

Custom connectors

You can add a custom connector by entering a remote MCP server URL under Customize and then Connectors. Two facts from Anthropic matter here.

1. Connections come from Anthropic's cloud, not from your device. The server has to be reachable over the public internet, so a server behind a firewall or on a private network may fail to connect.
2. Anthropic has not verified custom connector services. Only connect to servers from organizations you trust, and review their authentication permissions.

Free users are limited to one custom connector. A custom connector that uses a shared credential can reach whatever that credential can reach, which can be far more than any single person's access.

Where your data goes

Anthropic's page says data transfers are encrypted, and that you can only sync content you have permission to view in the source. It also says third-party services process data under their own terms, which may be outside the United States. For Team and Enterprise, connectors are only available in private projects, and chats with synced content cannot be shared.

One related point from Anthropic's privacy page for consumer plans: raw content from connectors, such as Google Drive files, is excluded from model improvement, but it can still be included if you copy it directly into a chat. The line between "Claude read it through a connector" and "I pasted it" matters.

Five questions before you connect anything

A first connector that earns its place

Pick a source you search often and one clear question you ask it. A drive you search for the latest version of a document, or a calendar you ask about next week. Try it with a question whose answer you already know, so you can see what Claude finds and what it misses. After a week, decide whether it saved time. If not, disconnect it.

Other assistants have their own versions of this, with their own limits: Gemini Connected Apps: What It Actually Shares, Setting Up Grok Connectors for GitHub, Notion and Google and Setting Up Copilot Connectors for OneDrive and Outlook cover them.

Official sources

• Pre-built integrations using remote MCP | Claude Help Center
• Is my data used for model training? | Claude Privacy Center

Checked on October 24, 2026. Features, plans and names change often, so the vendor's own pages are the final word.