Skip to content

How to add a custom connector to Claude Team or Enterprise

Step-by-step admin guide to adding a custom connector to a Claude Team or Enterprise organization: who can add it, the settings to choose, how members connect, and a rollout checklist.

Updated

Who can add a connector

Anthropic documents that anyone can build and host a connector, but adding one to an organization is restricted. On Team plans only Owners and Primary Owners can add custom connectors. On Enterprise plans, Owners, Primary Owners, and people with a custom role that includes Manage access to Libraries can add them.

Adding a connector to the organization does not switch it on for everyone. Members connect to it and enable it themselves. Anthropic describes this as the way Claude stays limited to the tools and data each user can already access, unless the connector uses a shared credential such as an API key.

Before you start

Have these ready before an owner opens the settings page:

  • The remote MCP server URL, on a public HTTPS address Claude can reach from Anthropic's cloud. If it sits on a private network, allowlist Anthropic's IP ranges first.
  • The authentication method the server uses: OAuth, an API key or other fixed credential, or none.
  • If OAuth, whether your authorization server accepts Claude's published identity, automatic registration, or needs its own OAuth client ID and secret.
  • A clear name users will recognize in their connector list.
  • A short list of what the connector can do, including any write actions.

Admin steps

These are Anthropic's documented steps for Team and Enterprise owners at the time of writing:

  • Go to Organization settings, then Connectors.
  • Click Add, hover over Custom, and select Web.
  • Enter the connector name users will see.
  • Enter the remote MCP server URL and click Continue.
  • Review the authentication settings Claude detected, adjust them if needed, and click Continue.
  • Under Authentication, choose Sign in now, Sign in when needed, or No sign in.
  • Under OAuth client, choose Use Claude's published identity (Anthropic's recommended option), Register automatically, or Use your own OAuth client. This setting is hidden if you chose No sign in.
  • Under Request headers, add any fixed credential the server needs, such as an API key.
  • Click Add.

Per-user authentication versus a shared credential

This is the most important choice in the whole setup. With OAuth sign-in, each member authenticates as themselves, and Anthropic notes that Claude inherits each person's permissions from the connected service. If someone cannot open a file or record in the source system, the connector cannot reach it from Claude either.

With a fixed credential in Request headers, Anthropic is explicit that everyone who uses the connector reaches the service with that same credential. If you must use one, create a credential limited to what every member should be allowed to see, and never use an administrator key.

What members do

Members can disconnect the connector at any time from their own settings, and owners can remove it for the whole organization under Organization settings, then Connectors. Tell users both, so they know access is under control.

Once the connector is added, members go to Customize, then Connectors, find the connector (it usually carries a Custom label), and click Connect to sign in. In a chat, they enable it with the plus button, then Connectors, using the toggle for each conversation.

If the connector URL is on a domain that matches a Connectors Directory listing, Anthropic says it may show that service's name and branding instead of the Custom label. It is still your connector and uses the exact URL the owner entered.

Rollout checklist

  • Test with an owner account and a regular member account, and confirm each sees only their own data.
  • Start with read-only tools. Add write tools only after a pilot group has used it.
  • Pick five to ten pilot users with real questions, and collect the questions Claude answered badly.
  • Tell users how to review tool approval prompts, and to choose Allow always only for tools they trust to run unsupervised.
  • If people use Research, disable write tools for it, since Anthropic notes Research can call connector tools without further approval.
  • Write a one-page guide: what the connector does, how to connect, and who to ask.
  • Remember that editing a connector means removing it and adding it again, so plan URL or auth changes for a quiet window.

Common mistakes, and having it done for you

Most rollout problems come from setup choices that seemed harmless at the time. A few are worth checking before the connector reaches the whole organization.

Adding the connector takes minutes. The work around it, choosing the auth model, testing permissions with real accounts, and running the pilot, takes longer. With Prestyj's done-for-you Claude connectors, we handle that rollout in your workspace along with the server itself, and leave your team with the user guide.

  • Testing only as an owner. Owners often have wider access in the source system, so the connector looks better than it will for members.
  • Using an administrator API key in Request headers. Every member then has administrator reach through Claude.
  • Pointing at a server only reachable on the office network. Claude calls from Anthropic's cloud, so it will fail for everyone.
  • Launching with write tools switched on before anyone has seen how Claude uses the read tools.
  • Giving the connector a technical name members do not recognize, so nobody turns it on.
  • Skipping the user guide. People who do not know a connector exists, or when to enable it, will not use it.

Sources

Plan rules and product names from Anthropic and OpenAI change often. Check the linked documentation for your plan before you plan a rollout.

Want this built for your systems?

Tell us which systems Claude or ChatGPT should reach and who should see what. We will scope it in writing.

Book a scoping call