Claude connectors: directory, custom, plans and teams

Updated 9 min read

Claude connectors link Claude to outside apps such as Google Drive, Slack, Linear or your own systems, so it can read from them and act in them. Each connector is a remote MCP server: either one listed in Anthropic’s Connectors Directory after review, or a custom connector you add by its URL. Custom connectors work on every plan, and the Free plan allows one. On Team and Enterprise, an Owner adds a connector for the organization and each member then signs in with their own account.

Menus and plan rules below match Anthropic’s help center and docs on 12 October 2026.

What is a Claude connector?

A connector is Claude’s name for an MCP server added to your account. MCP (Model Context Protocol) is the open standard Anthropic created for AI apps to reach tools and data. Directory and custom connectors share the same runtime and tool-calling code; they differ only in review, discovery and distribution.

Remote connectors are brokered through your Claude account, so Claude reaches the server from Anthropic’s cloud. That holds even in Claude Desktop and Cowork, which run on your computer. A server behind a VPN or corporate firewall will not connect, even if you can open it yourself, until you allowlist Anthropic’s IP ranges.

Where each kind works, from Anthropic’s getting-started page and help center:

  • claude.ai, the mobile apps and Cowork: remote connectors.
  • Claude Desktop: remote connectors, plus local desktop extensions (MCP Bundles). Local servers in claude_desktop_config.json use your own network and are not available in claude.ai or Cowork.
  • Claude Code: the connectors you add in claude.ai, when you sign in with your Claude account.

So the answer to “Claude connectors vs MCP” is that a connector is MCP. In Claude Code you can also add servers directly with claude mcp add, which our Claude Code MCP guide covers.

Directory connectors or custom connectors?

Directory connector Custom connector
Anthropic review Yes No
Found inside Claude Browse, search, Suggested Connectors No
How people install it Listing URL claude.ai/directory/connectors/<slug> Install link, or the URL typed in
Anthropic-held OAuth client Available Not available
Shown as Named card with logo Custom

Directory listings are labelled Verified (Anthropic tested the tools, which it says is not a security audit) or Community (screened without an in-depth review; Claude shows a reminder before you connect). Anything you add by URL shows as Custom.

On 12 October 2026 the Claude marketplace listed 937 connectors and 342 plugins. Its top connectors were Google Drive, Gmail, Google Calendar, Canva, Microsoft 365, Notion, Figma, Slack, HubSpot, Asana, Linear and monday.com, all marked Anthropic verified. For engineering teams, Atlassian MCP (Jira, Confluence, Bitbucket), Supabase and Grafana Cloud are listed too.

Only a directory listing makes a connector discoverable inside Claude. The open MCP Registry does not, and custom connectors are never suggested in chat.

Which Claude plans include connectors?

Custom connectors using remote MCP are available on Free, Pro, Max, Team and Enterprise, in Claude, Cowork and Claude Desktop. Free accounts are limited to one custom connector.

Plan Custom connectors Who adds them
Free One You
Pro, Max Yes You
Team Yes Owners and Primary Owners
Enterprise Yes Owners, Primary Owners, and custom roles with Manage access to Libraries

For directory connectors, Anthropic’s getting-started page asks only for a Claude account and an account with the service. That service may need its own paid plan. What a Team seat costs and includes is in our Claude Team plan guide.

How do I add a connector in Claude?

From the directory:

  1. Open Customize > Connectors at claude.ai. In the desktop app, select Customize in the sidebar, then Connectors.
  2. Select Discover, search for the service, and open its page.
  3. Select Connect to Claude, sign in to the service, and approve the access it asks for.
  4. The connector appears under Your connectors as Connected.

A custom connector on Pro or Max:

  1. Open Customize > Connectors.
  2. Click + Add, then Add custom connector.
  3. Enter a name and the server’s remote MCP URL, then click Continue.
  4. Review the authentication settings Claude detected and click Continue.
  5. Under Authentication, pick Sign in now, Sign in when needed, or No sign in. Under OAuth client, pick Use Claude’s published identity (Recommended), Register automatically, or Use your own OAuth client. Under Request headers, add a fixed API key if the server takes one instead of OAuth.
  6. Click Add.

You then turn connectors on per conversation: click + in the message box, then Connectors, and use the toggle.

You cannot edit a custom connector. Remove it from Customize > Connectors and add it again with the new details.

A server outside the directory can link straight to the Add custom connector dialog with the name and URL filled in:

https://claude.ai/customize/connectors?modal=add-custom-connector&connectorName=NAME&connectorUrl=ENCODED_URL

connectorUrl must be percent-encoded. Claude says the values came from an external link, and the person confirms them. The link only fills the form and grants nothing. Owners can use the same parameters on https://claude.ai/admin-settings/connectors to fill in the organization-wide dialog.

How do connectors work for a team?

On Team and Enterprise, adding a connector and connecting to it are separate steps.

An Owner adds a custom connector under Organization settings > Connectors: click Add, hover over Custom, select Web, then enter the name, URL and the same authentication settings as above. For a directory connector, the Owner’s button reads Connect for your team; the arrow next to it offers Connect just for me, to try it first. Members see Request for a connector the organization has not added.

Each member then opens Customize > Connectors, finds the connector (custom ones usually carry a Custom label) and clicks Connect. With OAuth, each person signs in as themselves, so Claude reaches only what that person can reach in the service.

The exception is a shared credential. If the Owner enters an API key under Request headers, everyone who uses the connector reaches the service with that one key, so Anthropic advises a key limited to what members should access. Anthropic’s docs call request headers a beta for a limited set of organizations.

A workable rule: per-person OAuth for anything with per-user permissions (issue trackers, documents, mail), and shared keys only for read-only or narrowly scoped services.

Are Claude connectors secure?

They are as safe as the server you connect and the permissions you grant. Anthropic’s guidance:

  • Custom connectors are not verified by Anthropic, and Anthropic does not run third-party servers or control how they handle your data. Connect only to developers you trust.
  • With OAuth, Claude never sees your password. Review the scopes the service asks for, and deny access if they look broader than the job needs.
  • A malicious server can carry hidden instructions (prompt injection), and a developer can change a server’s tools at any time. Watch tool inputs, outputs and behavior.

You decide what Claude may do with each connector. On its page under Customize > Connectors, Tool permissions sets each tool or group of tools to Always allow, Needs approval or Blocked. In chat, choose Always allow only for tools you trust to run unsupervised. During Research, Claude can call connector tools without asking again, so turn off tools that write before you start one.

To revoke access, select Disconnect on the connector’s page, or revoke Claude in the service’s settings.

What happens on the server when Claude connects?

We built a remote MCP server that Claude uses as a custom connector, and signed in through it for real on 28 September 2026. The server’s side of the flow shows what “Claude acts as you” means. In order, with Anthropic’s requirements from its authentication docs:

  1. Claude calls the MCP URL with no token. The server must answer 401 with a WWW-Authenticate: Bearer resource_metadata="..." header pointing at its protected resource metadata. Claude ignores that header on a 200.
  2. Claude reads that document, takes the first authorization server it lists, and reads that server’s metadata.
  3. Claude registers itself as an OAuth client through dynamic client registration (RFC 7591), with the callback https://claude.ai/api/mcp/auth_callback. Per the docs, Claude uses a Client ID Metadata Document (CIMD) instead only when the authorization server advertises support for it.
  4. The person’s browser opens the server’s consent page, with a PKCE S256 challenge. On ours they sign in, pick a workspace and click Allow access.
  5. The browser returns to the callback, and Claude exchanges the code for a token.
  6. From then on every tool call carries that person’s token, and the server acts as that person: in the workspace they picked, or another of theirs that a call names.

If your team builds its own server, accept that callback plus a loopback redirect on any port (Claude Code’s), and answer token requests within 10 seconds and refreshes within 30. DCR registers a new client on every fresh connection, so for heavy traffic the docs suggest CIMD. If your MCP server sees requests but your authorization server sees none, discovery failed, and Claude reports “Couldn’t reach the MCP server.”

Which connectors does a developer team need?

Start from where the team’s work already lives: the issue tracker (Linear, or Atlassian MCP for Jira and Confluence), documents (Notion, Google Drive or Microsoft 365) and chat (Slack). Each one reads live tickets, specs or messages straight from its own service. Our Notion MCP and Atlassian MCP guides go deeper on those two.

None of them holds what your agents have learned: how a system works, what was decided and why. Claude’s memory keeps context about you in your Claude account (see the Claude memory guide), and Claude Code, Codex and Cursor each keep their own. A shared wiki gives all of them the same pages.

Add Dexio as a Claude connector

Dexio is a hosted wiki that AI agents read and write through MCP, and people see what their agents know as a page graph at app.dexio.wiki. It is open source (AGPL-3.0), free for one person, and $10 a member a month on Team or $20 on Business.

It connects to Claude as a custom connector at https://app.dexio.wiki/mcp. While signed in to Claude, open this install link:

https://claude.ai/customize/connectors?modal=add-custom-connector&connectorName=Dexio&connectorUrl=https%3A%2F%2Fapp.dexio.wiki%2Fmcp

Click Add, then Connect, then Allow access on Dexio’s page, where you sign in (or create a free account) and pick the workspace. In a chat, turn Dexio on under +, then Connectors. On Team or Enterprise, an Owner adds Dexio once under Organization settings > Connectors, and each member clicks Connect and signs in as themselves.

Claude Code, Codex, Cursor and other MCP clients connect to the same address with an API key, so a decision worked out in a Claude chat is on the page Claude Code reads next session. page_history shows who changed a page, when and why. If you are weighing other tools for this, best MCP memory servers compares them.

FAQ

Are Claude connectors free?

A Free plan can add one custom connector, and directory connectors need only a Claude account. The service behind a connector may need its own paid plan.

Do Claude connectors work in Claude Code?

Yes. Connectors you add in claude.ai are available in Claude Code when you sign in with your Claude account.

Why can’t Claude reach my MCP server?

Claude connects from Anthropic’s cloud, so the server must be public or allowlist Anthropic’s IP ranges. If it is reachable, check that it answers 401 with a resource_metadata pointer.

Sources

Ask a question

Ask anything about Dexio.

About
We reply by email.