Claude connectors: what they are and how to use them

A Claude connector links Claude to an outside app or service, Gmail, Slack, Notion, Linear, so it can search, read, and sometimes act inside that service instead of just talking about it. Connectors ship on every plan now, from Free through Enterprise, and Claude can only reach what you could already see or do yourself in that app.

What a connector actually does

Per Anthropic's own Help Center, a connector lets Claude "access your apps and services, retrieve your data, and take actions within connected services," working the same way across Claude.ai, Claude Desktop, Claude Code, and the API.

There are two flavors. Standard connectors are pre-built and listed in Anthropic's Connectors Directory, reviewed before they go live, available to everyone. Custom connectors point Claude at a remote MCP server URL of your own choosing, a service Anthropic hasn't reviewed. Same mechanism, different trust level.

Connector vs. MCP: the part people mix up

MCP, the Model Context Protocol, is the open standard underneath all of this. Anthropic announced MCP in November 2024 as a way to build "secure, two-way connections between data sources and AI-powered tools," and it's since spread well past Claude: ChatGPT, Cursor, Gemini, Microsoft Copilot, and VS Code all support it now. In December 2025, Anthropic donated MCP's governance to a new Linux Foundation body, the Agentic AI Foundation, co-founded with Block and OpenAI, reporting "over 97 million monthly SDK downloads" and "10,000 active" public servers at the time.

A connector is Anthropic's product wrapper around that protocol: click to authenticate, no configuration, no server address to type in. MCP is the plumbing any AI tool-builder can use; a connector is what you actually click. Directory connectors are Anthropic-reviewed implementations of MCP servers. Custom connectors let you skip the review and point at any MCP server URL directly, which is exactly where the security section below applies.

Notable connectors, first-party and directory

Google Workspace and Microsoft 365 are Anthropic's own, with dedicated setup docs separate from the general directory. Google Workspace covers Gmail (search, read, draft, but not send), Google Calendar (view, create, modify events), and Google Drive (search, read, upload, create folders), available to all users on Claude and Claude Desktop as of Anthropic's most recent update to that doc, May 2026. Microsoft 365 covers SharePoint, OneDrive, Outlook mail and calendar, and Teams chat, read-only, with data staying inside your tenant.

The third-party Connectors Directory (claude.ai/directory) launched in July 2025 with Notion, Canva, Stripe, Figma, Socket, and Prisma as founding partners. Anthropic's own count, as of an April 2026 blog post, put the directory at "over 200 connectors, spanning popular apps for design, finance, productivity, and health," naming Linear, Slack, Spotify, Uber, Instacart, and Booking.com among others. You'll find bigger numbers, 343, 375+, even 554, floating around third-party trackers and SEO blogs, but those disagree with each other and aren't Anthropic sources, so treat "200+" as the reliable figure and assume the real count has kept growing since.

What's free, and how to add one

Every plan gets connectors. The gate is on custom ones: Free-plan users are capped at one custom (remote MCP) connector, per the Help Center article on custom connectors. Pro, Max, Team, and Enterprise aren't stated to carry that cap. Standard directory connectors are open to every plan regardless, a change from the July 2025 launch, when remote third-party connectors were paid-only. On Team and Enterprise, only an Owner can enable a connector org-wide; individual members then authenticate themselves, and Owners can restrict scopes, disable specific actions, or limit connections to approved domains.

The security model

Connectors run on OAuth, so Claude authenticates through the service's own login flow and never stores your password. The part worth actually internalizing: Claude inherits your existing permissions in the source system. Anthropic states it plainly for the general case: "if someone can't access a specific file, channel, or record in the source system, the connector can't reach it from Claude either," and confirms the same for Google Workspace specifically, "Claude mirrors your existing permissions."

That's a real ceiling, not marketing language, but it isn't the whole picture. A connector granted write scope can genuinely take actions on your behalf, sending a Slack message, filing a Linear issue, so the scope prompt at authorization time is the moment that matters, not something to click through.

Directory connectors go through an Anthropic review before listing: tool design, authentication, privacy, allowed external links, documentation, and how the connector behaves when every tool gets called. Custom connectors skip that review entirely, which is why Anthropic's own guidance is blunt about it.

Tested on Claude Sonnet 5, July 2026: connecting Gmail through the directory required the standard Google OAuth screen, listing read/draft/label scopes explicitly before granting access, and Claude was unable to find messages outside what that Google account itself had access to, consistent with the documented permissions model.

This site isn't run by Anthropic. We're an independent guide to using Claude well, and where the official numbers conflict, like the directory's exact connector count, we cite Anthropic's own figure rather than repeating whichever third-party total sounds biggest.

Are Claude connectors free?

Yes, on every plan including Free. The one real limit is custom connectors: Free-plan accounts are capped at one, while Pro, Max, Team, and Enterprise aren't stated to have that cap. Standard directory connectors, Gmail, Slack, Notion, and the rest, are open to all plans with no extra charge.

What is the difference between a connector and MCP?

MCP is the open protocol, the actual technical standard for connecting an AI system to outside data and tools. A connector is Anthropic's product layer on top of it: a listing you click to authenticate, with no server URL or configuration required. Every connector is built on MCP, but not every MCP server is a listed connector, custom connectors let you point at one directly.

Are Claude connectors safe?

Standard directory connectors go through an Anthropic review before listing, covering tool design, authentication, and privacy. Custom connectors don't get that review, so Anthropic's own advice is to only add ones from sources you trust. In both cases, Claude can only reach what you could already access in that service, it can't escalate beyond your existing permissions.

Do I need Claude Code to use connectors?

No. Connectors work the same across Claude.ai, Claude Desktop, and the API, not just Claude Code. If you're setting up Claude for the first time, how to use Claude walks through your first session, and Claude Projects and Claude Skills cover the other two pieces people mix up with connectors. For the wider set of write-ups, browse more reviews.