Claude connectors are ready-made integrations that connect Claude to your apps in claude.ai, the desktop app, mobile and Cowork: search files in Google Drive, summarize threads in Gmail, send a message in Slack, update a page in Notion. Technically, almost every connector is a remote MCP server you add from Anthropic's directory with a click and sign in to with your own account; only desktop extensions run locally. Claude only sees what you are allowed to see in the service itself. According to Anthropic, directory connectors are open to all users, including the free plan; custom connectors by URL exist on every plan, but Free gets only one. On Team and Enterprise, an owner first enables a connector for the organization, then each person signs in individually. For each tool you decide whether Claude may always use it, must ask first, or can't use it. What connectors can't do: grant rights you don't have in the service, reach a server on your private network as a custom connector, or read the content of Gmail attachments.
I work with Claude every day and run my team of fourteen AI employees on it. This article explains connectors based on Anthropic's help center and docs as of November 1, 2026. Everything about Claude is collected on my Claude page. To set up MCP servers yourself with a command in Claude Code, read Set Up a Claude MCP Server; this article is about the click-to-add Claude integrations in claude.ai.
What Claude connectors are
According to Anthropic, connectors let Claude access your apps, retrieve your data and take actions. A connector inherits each person's permissions from the connected service. If you can't open a file in Google Drive, neither can Claude.
You find connectors in the directory, Anthropic's catalog. Every listing has a page with use cases, read and write capabilities and availability, plus a "Verified" or "Community" label. On the Claude Marketplace website, Anthropic lists over 2,000 connectors and plugins combined.
There are two kinds:
- Remote connectors: cloud services you sign in to, such as Slack, Notion or GitHub. According to Anthropic they work everywhere you use Claude: web, desktop, mobile, Cowork and Claude Code.
- Desktop extensions: local connectors for files, databases on
localhost or desktop applications. They only run in Claude Desktop and Claude Code, not on the web or mobile.
If you sign in to Claude Code with your claude.ai account, your connectors are available there automatically, marked as such in /mcp; with an API key or Amazon Bedrock, Claude Code doesn't load them, according to the docs. How connectors, MCP servers, skills and plugins fit together is covered in Claude Code Extensions: Plugins, Mods, Skills.
New: connectors on the Plugins page
Since October 29, 2026, Anthropic has been gradually merging plugins, skills and connectors onto one page. The "Customize" section in the sidebar becomes "Plugins", and one page replaces the separate Skills, Connectors and Plugins tabs. Quick filters narrow the list to skills or connectors, and each card shows its type. How skills work is explained in Claude Skills: How to Create and Use Them.
Your connectors stay installed, according to Anthropic. One trap: removing a plugin that bundles skills and a connector leaves the connector connected; you disconnect it separately. How plugins work otherwise is covered in Claude Code Plugins: Install, Build, Trust.
Which connectors exist
On November 1, 2026, the Claude Marketplace listed 1,050 connectors. The most popular ones, with Anthropic's short descriptions:
Which plans include connectors
For Microsoft 365 you need a Microsoft Entra tenant tied to a Microsoft Business plan, according to Anthropic; personal accounts such as @outlook.com don't work. An Entra administrator has to approve the integration once for the tenant. For Outlook and my calendar I don't use this connector myself but an MCP server. What both paths mean compared with Copilot is covered in Copilot vs Claude: What Works at the Office.
Enabling connectors on Team and Enterprise
An owner or primary owner enables a connector under Organization settings > Connectors ("Add to your team"). That doesn't give anyone access yet: each person then signs in with their own account and only sees their own data. The exception is custom connectors with a shared credential: then everyone who uses the connector reaches whatever that credential can access. Whether Team pays off for you is covered in Claude Team: Cost and When It Pays Off.
If a connector is missing, members can ask for it with "Request". On Team plans this is on by default, on Enterprise plans off. There is also a beta for signing in once for the whole organization; on Enterprise, custom roles control which connectors and tools a group may use. What Enterprise adds on top is compared in Claude Enterprise or Team: Cost and GDPR.
Permissions: read, write, ask first
For each group of tools, or for a single tool, you pick one of three levels under "Tool permissions":
- Always allow: Claude uses the tool without asking.
- Needs approval: Claude asks before every call. You answer with "Allow once" or "Always allow".
- Blocked: Claude can't use the tool.
Tools are grouped by type, such as read-only and write/delete, so you can let Claude search email but not send it. On Team and Enterprise, owners set these limits for the whole organization, and individual members can't override them. A permission in Claude never grants more than the service itself allows; it can only narrow it.
For Gmail and Google Drive, Claude asks by default before sending, replying to or forwarding email and before sharing, moving or trashing files, according to Anthropic. On Team and Enterprise, owners decide whether members may turn that off. In each conversation you can also switch every connector on or off under "+" and "Connectors". With ten or more active connectors, Anthropic suggests the "On demand" setting to give your conversations more room.
Privacy: where your data ends up
All data transfers are encrypted, according to Anthropic. Three points are worth knowing:
- The service stays in charge. Connected services process data on their own infrastructure, under their own terms, possibly outside the United States. Settings that control where Claude runs don't change that.
- Retrieved data stays with the chat. For the Google connectors, Anthropic stores retrieved data with its chat; delete the chat and the data is gone. Anthropic says it doesn't train on Gmail, Drive or Calendar connector data.
- Training depends on the plan. On Free, Pro, Max and Ultra, raw content from connectors isn't used for training, even if you have allowed training; what you copy into the chat yourself can be. On commercial plans, Anthropic doesn't use inputs and outputs for training by default.
On Team and Enterprise, connectors only work in private projects, and chats with synced content can't be shared. How projects work is covered in How to Use Claude Projects Right. What applies legally to Claude is covered in Claude Privacy and GDPR: What Applies. What an AI should see when customer data is involved is covered in AI and Privacy: What the AI Gets to See.
Adding a custom connector by URL
If a service isn't in the directory but has a remote MCP server, you add it as a custom connector:
- Open Customize > Connectors (Plugins after the switch), choose "Add", then "Custom" and "Web". On Team and Enterprise, an owner does this under Organization settings > Connectors.
- Enter a name and the MCP server's URL.
- Choose how people sign in: "Sign in now", "Sign in when needed" or "No sign in". If the server uses a fixed key instead of OAuth, add it as a request header.
- Save with "Add" and connect it like any other connector.
Important: Claude connects to your server from Anthropic's cloud, not from your computer, even in the desktop app and in Cowork. So the server must be publicly reachable. If it sits behind a firewall, you need to allowlist Anthropic's IP ranges. Anthropic hasn't verified custom connectors. The help center warns about hidden instructions in malicious servers (prompt injection) and about tools that change with an update. More on securing Claude Code is in Claude Code Security: Permissions and Sandbox.
Limits: what connectors can't do
- No access to private networks: custom connectors must be reachable from the internet. Local MCP servers from Claude Desktop's config file aren't available on the web or in Cowork.
- Gmail attachments: metadata only, not the content.
- Google Drive: Claude reads the text only, not embedded images.
- Microsoft 365: write tools, meaning sending email, calendar events, files and Teams messages, only work once an Entra administrator has approved them. They can't send attachments, and Claude can't change Teams settings or permissions.
- HIPAA: if an organization uses the HIPAA configuration, the BAA doesn't cover connectors, plugins or skills.
Frequently asked questions
What is the difference between a connector and an MCP server?
A connector is an MCP server you add to your claude.ai account instead of configuring it with a command in Claude Code. It then works wherever you are signed in with that account.
Do Claude connectors cost extra?
Not from Anthropic; connectors work on every plan. The connected service may require its own account or a paid plan, though. What the plans themselves cost is in Claude Pricing and Plans Explained.
Can Claude send emails through a connector?
Yes, if you allow it. With Gmail, Claude asks before every send by default. To rule it out completely, set the write tools to "Blocked".
Does the new Plugins page change my connectors?
No. According to Anthropic they stay connected; only their place in the menu changes.
Where to go from here
Start with one connector you need every day, and set its write tools to "Needs approval" first. New to Claude? Start with Claude Guide: Getting Started; which learning options exist beyond that is compared in Claude Courses: What Exists, What's Worth It. For bigger tasks across several apps, read Claude Cowork: What It Does, Who It's For. Which integrations hold up in practice is what we discuss in my English-language community, the AI Employee Framework.
Your first AI employee up and running within an hour
Join the community$5/month founding price, first 50 members
