What Are Claude Connectors? Directory vs Custom Connectors Explained

A Claude connector is a link between Claude and an outside app or service, built on the Model Context Protocol (MCP). Most connectors are remote MCP servers that Claude reaches over the internet. There are two kinds: directory connectors, which Anthropic lists after a check, and custom connectors, which a user or admin adds by pasting a server URL.

Both kinds run the same way once connected. The difference is review, discovery and distribution. This guide explains the two, which Claude plans can add them, and where they show up, all checked against Anthropic's documentation on 6 November 2026.

If you build software, this matters because a connector is how your product shows up inside Claude. If you are deciding whether to get listed, read our step-by-step guide on how to get listed in the Claude connectors directory after this one.

What a Claude connector is

Anthropic's getting started page says a connector "links Claude to an outside app or service, such as your calendar, your documents, or your issue tracker, so Claude can find information there and take actions for you."

Under the hood, a connector is an MCP server. MCP is the open standard that lets AI apps call tools in other systems. For Claude, the server exposes tools, Claude decides when to call them, and the user approves calls according to the connector's tool permissions.

Most connectors are remote. The same page says "Most connectors are remote services that you sign in to." A local connector is a desktop extension packaged as an MCP Bundle (MCPB) that runs on your computer inside Claude Desktop. Our explainer on MCP desktop extensions covers that format.

Directory connectors vs custom connectors

Anthropic's directory vs custom comparison puts it plainly: "Both run on the same MCP infrastructure: the runtime, transport, authentication, and tool-calling code paths are identical. The difference is review, discoverability, and distribution."

Source: Anthropic's directory vs custom page, checked on 6 November 2026.

Directory connectors

The Connectors Directory page describes "a catalog of MCP servers that work across all Claude products." Two points stand out for vendors:

• Directory connectors are eligible for Suggested Connectors, "in-chat recommendations shown when relevant to the user's task." Every directory entry is included automatically.
• "Ranking is usage-based, similar to other app stores."

Each listing carries a label. Anthropic's verification page says a Verified connector has had its tools tested by Anthropic "for quality and compatibility" and shows a checkmark. A Community connector was screened "but hasn't reviewed them in depth," and Claude shows users a reminder before they connect it. Anthropic adds that verification "isn't a security audit."

Custom connectors

A custom connector is any remote MCP server added by its URL. Anthropic's custom connector page says "You can add any third-party connector as a custom connector as long as you have the URL of its remote MCP server." It also warns that custom connectors "allow connections to unverified services."

Vendors use custom connectors in three common ways. They test their own server before submitting it. They give enterprise customers a separate URL. And they ship before a listing is approved. Anthropic's comparison page names the second one as "a supported pattern": list a connector with safe defaults, and give enterprise customers a separate URL "with elevated permissions or tenant-specific configuration."

You can also send users a link that opens the "Add custom connector" dialog with your name and URL filled in. The link uses the modal=add-custom-connector, connectorName and connectorUrl parameters. Anthropic notes that install links "only prefill the form" and the user still confirms.

Which Claude plans can add connectors

Plan rules differ between custom connectors and directory submissions, so keep them apart.

On Team plans, members without permission see a Request button on directory connectors. The request goes to the organization's Owners, per the directory page.

Where connectors appear

The directory page says "The same catalog serves claude.ai, Cowork, Desktop, mobile, and Claude Code." The getting started page breaks it down by app:

• claude.ai on the web: remote connectors, including MCP Apps that show interactive content in the conversation.
• Claude Desktop: remote connectors, plus local connectors installed as desktop extensions.
• Claude mobile app: remote connectors.
• Claude Code: connectors added in claude.ai are available when you sign in with your Claude account.
• Cowork: connectors and plugins.

Users add and manage connectors under Customize > Connectors. In a chat, they switch each connector on or off from the + menu under Connectors.

Claude Code picks up claude.ai connectors

Claude Code's MCP documentation says that if you log in with a claude.ai account, "MCP servers you've added in claude.ai, known as connectors, are automatically available in Claude Code." They are fetched only when the active login is a claude.ai subscription. They are not loaded when an API key, a third-party provider such as Amazon Bedrock, or certain token setups are active. Users can turn them off with the disableClaudeAiConnectors setting or ENABLE_CLAUDEAI_MCP_SERVERS=false.

For your product, this means one remote server can reach a user in chat, in the desktop app and in their terminal. Our comparison of Claude Desktop vs Claude Code for MCP covers the differences for users.

Remote MCP under the hood

Because connectors are MCP servers, the engineering work is building a solid remote server. Anthropic's authentication page says the same authentication infrastructure "backs claude.ai, Claude Desktop, Claude mobile, Claude Code, and Cowork." A few requirements from that page:

• Supported sign-in types include OAuth 2.0 with Dynamic Client Registration, OAuth with a Client ID Metadata Document, OAuth with Anthropic-held client credentials, and no authentication.
• The hosted Claude apps use one redirect URI: https://claude.ai/api/mcp/auth_callback. Claude Code uses a loopback redirect on a varying port.
• Claude includes a PKCE code_challenge with S256 on every authorization request.
• Anthropic's traffic to your server comes from 160.79.104.0/21.

A static API key in a request header exists too, but Anthropic marks it as beta "for a limited set of organizations." If each person needs their own account, Anthropic says to use OAuth. Our tutorial on adding OAuth to a remote MCP server walks through the server side.

Directory listing vs the MCP Registry

Vendors often assume one listing covers everything. It does not. Anthropic's comparison page says "The Anthropic Directory is independent of the open MCP Registry" and "Publishing to those doesn't surface your server in Claude." The directory page adds that no DNS or .well-known domain proof is required for the Anthropic directory, since "That requirement applies only to the open MCP Registry."

So most MCP companies do both. Our explainer on the MCP Registry covers the open registry, and where to sell or list your MCP server maps the other channels.

What this means for a software company

If your customers use Claude, a connector is your product's front door there. A practical sequence looks like this:

1. Build or harden a remote MCP server with OAuth.
2. Test it as a custom connector in your own Claude account.
3. Publish an install link in your docs so customers can add it now.
4. Submit it to the directory for search, a named card and Suggested Connectors.
5. Keep a separate enterprise URL if some customers need different permissions.

Each step has its own requirements. The Claude connector directory listing guide covers step 4 in detail.

Next step

If you want a remote MCP server that works as a Claude custom connector and is ready for directory review, see eSEOspace's Claude connector development service. We scope each project individually. For the wider launch plan, the MCP launch hub collects the related guides.

Frequently asked questions