Safeguard
AI Security•

Who can add a custom connector on Claude Team and Enterprise?

On Claude Team only Owners add custom connectors, and Enterprise adds custom roles with library access. See who decides, who connects, and what to approve.

Safeguard Security Team
7 min read

On Team plans, only Owners and Primary Owners can add a custom connector for the organization. On Enterprise, Owners, Primary Owners and custom roles with Manage access to Libraries can. Members then connect individually. On Pro and Max, the account holder adds it. That makes connector approval a governance decision, not a personal one.

Who adds what

Claude's help center for custom connectors gives the rules by plan.

PlanWho can add a connector
FreeThe user, limited to one connector
Pro and MaxThe individual account holder
TeamOwners and Primary Owners
EnterpriseOwners, Primary Owners, and custom roles with Manage access to Libraries

After an admin adds a connector on Team or Enterprise, each member still connects individually under the connectors settings. So adding and authenticating are two separate steps with two different people involved.

Why this matters for security tools

A security connector can expose vulnerability details, asset names and ticket data. Adding it is a data sharing choice. If any member could add one, the list of systems touching your findings would grow without review.

Restricting the add step to a few owners gives you a natural change control point. It also gives auditors a short answer to the question of who can introduce a new tool into the assistant.

A simple approval routine

  1. A requester names the connector, its owner, and the reason.
  2. Security reviews the tool list, the sign in method and the scopes.
  3. An Owner adds it, ideally with write tools disabled at first.
  4. Members connect with their own accounts.
  5. Review again after any change the vendor announces.

Settings worth knowing

The help center lists several controls that fit this routine. Unneeded tools can be disabled from the search and tools menu. Owners can disable interactive tool calls in organization connector settings. Tool approval prompts offer an Allow always choice, which should be used only for servers and tools you trust to run unsupervised.

It also warns that custom connectors are not verified by Anthropic and that server developers can change tool behavior without notice. Put a calendar reminder on connector reviews for that reason.

What to record

Keep a short register with the connector name, owner, date added, scopes, and the person who approved it. If your questionnaire asks who can add an AI integration, this register answers it.

Edge cases to plan for

People leave, roles change, and vendors rename things. Decide in advance who becomes the owner of a connector when its original requester moves on. Decide too what happens to members who connected under their own accounts: their sign in should be revoked at the vendor, not only removed from the chat plan.

Another case is a connector that exists on a personal Pro or Max account before the organization has a policy. The help center says the individual account holder adds connectors there, so a policy that only covers Team administrators will miss them. Ask people to declare any connector they use for work data, and keep that declaration in the same register.

A worked example: approving a security connector in a company of 200

This is a made up scenario to show the sequence.

A security engineer asks for a connector to a vulnerability tool. The request goes to the security lead. The lead reviews the tool list, finds twelve read tools and four write tools, and asks the vendor how write tools can be switched off. The vendor explains that administrators can disable them. The lead approves the connector with write tools off.

An Owner on the Team plan adds the connector. Then the lead sends a short message to the pilot group of six: connect with your own account, review the permissions you are asked to approve, and keep approval prompts on. Each person connects individually, as the help center describes.

Two weeks later the group reports that they keep asking for a ticket creation feature. The lead reviews the tool, tests it in a sandbox project, and approves one write tool with prompts on. The register is updated with the date and the reason.

Step by step: build the approval checklist

  1. Request. Name, vendor, address, owner, business reason.
  2. Data review. What categories of data can the tools return? Does any include customer data?
  3. Access review. Sign in method, scopes, shared credential or not.
  4. Tool review. Complete list with read or write label, and a test of each write tool in a sandbox.
  5. Control review. Approval prompts, tool disable options, owner settings for interactive tool calls.
  6. Decision. Approve, approve with limits, or decline, recorded with a name and a date.
  7. Re-review date. Because the help center warns that server developers can change tool behavior without notice, set a date, such as quarterly.

Who should be an Owner

The Team plan limits adding to Owners and Primary Owners, so these people are your gatekeepers. Keep the number small. Pick people who understand the data the connector can reach, not only the people who administer billing. If your security lead is not an Owner, agree on a handoff so the lead reviews before the Owner clicks add.

Table: responsibilities

RoleResponsibility
RequesterExplains the need and tests in a sandbox
Security reviewerReviews tools, scopes and data flow
Owner on the chat planAdds the connector after approval
Connector ownerWatches for vendor changes, rotates keys
MemberConnects with own account, reads prompts

What an auditor may ask

A reviewer may ask who can introduce a new integration into your AI assistant and how you know the list is complete. Your register, the plan's owner list and the approval records answer both. Do not claim controls you have not operated, and do not describe the register as an attestation of any kind. It is your own record.

When to say no

Decline when the operator cannot be named, when the connector wants write access for a read only purpose, or when the vendor will not state how tool changes are announced.

Where Safeguard fits

Safeguard offers asset discovery and an MCP Guardrail for teams that want inventory and policy around the tools agents call, which is the register above with some of the upkeep taken off your hands.

Try it yourself

Click through it at your own pace, or open it full screen.

Watch how a connected assistant gets its tools, then picture which of your administrators would need to approve that step.

Quick check

Can a regular member add a connector on a Team plan?

Not for the organization. The help center says only Owners and Primary Owners can add connectors to a Team organization.

Does adding a connector connect everyone automatically?

No. Each member connects individually after an admin adds it, so each person goes through their own sign in.

Does adding a connector give members access to everything it can reach?

No. Each member connects with their own account, and the connector can only reach what that sign in allows.

Sources

Last reviewed: 19 October 2026.

Never miss an update

Weekly insights on software supply chain security, delivered to your inbox.

Self-healing security runs on Safeguard.

Your first fix PR is minutes away.

No sales call required, even your agent can complete the purchase over MCP.