AI tool register · Claude
What can Claude see and do through connectors?
Claude connectors inherit the permissions of the person who connected the service, so they can’t reach what that person can’t. They can retrieve information and take actions, and each tool can be set to always allow, need approval, or be blocked. On Team and Enterprise, an owner has to enable a connector before anyone can use it.
Connectors (web connectors and desktop extensions)
Web connectors are available on all plans. Custom connectors: Free (limit of 1), Pro, Max, Team, and Enterprise.
- Google Drive
- Gmail
- Google Calendar
- Microsoft 365
- Slack
- GitHub
- Linear
What a connector can see, and what it can do
What a connector can see
Claude inherits each person's permissions in the connected service and cannot reach what that person cannot. A custom connector that uses a shared credential reaches whatever that credential can access.
Source (opens in a new tab)What it can do
Connectors can retrieve data and take actions, such as creating issues or sending messages. Each tool can be set to Always allow, Needs approval, or Blocked. Team and Enterprise owners can restrict actions for the whole organization.
Source (opens in a new tab)Who switches it on
Individual plans: each user connects their own services. Team and Enterprise: an Owner or Primary Owner must first enable a connector for the organization, then each person signs in individually.
Source (opens in a new tab)Whether that data trains the model
Anthropic says training and feedback data do not include raw content from connectors, including remote and local MCP servers, unless that content is copied directly into the conversation.
Source (opens in a new tab)Custom connectors
Yes, through remote MCP servers. Free (1 connector), Pro, Max, Team, Enterprise. On Team, only Owners and Primary Owners can add them. Anthropic does not verify custom connectors.
Source (opens in a new tab)
Our take
Before you connect anything
Set anything that sends or changes something to “needs approval”. Be careful with a custom connector that uses one shared login: it reaches whatever that login can, for everyone who uses it.
- Tidy permissions first. A connector can reach whatever the person who connected it can.
- Treat reading and acting as 2 separate decisions.
- Put the approved connectors in your AI use policy.
Compare and decide
- Claude: data handling by planDoes Claude train on your data?
- ChatGPT connectorsWhat can ChatGPT see and do once you connect your tools?
- Microsoft Copilot connectorsWhat can Microsoft Copilot see and do through connectors?
- Google Gemini connectorsWhat can Gemini see and do in your Google Workspace?
- What can AI see when I connect a tool?The general answer, without the product names.
- AI agentsWhat comes after connecting: an agent with a job.
How this was checked
Every fact on this page is taken from Anthropic’s own documentation as it read on . Vendors change these terms often, sometimes monthly. Check the linked page before you rely on it.
- Claude pricing comparison table (model training row, admin controls per plan) (opens in a new tab)
- Is my data used for model training? Consumer plans (opens in a new tab)
- Is my data used for model training? Commercial plans (opens in a new tab)
- How long do you store my data? Consumer plans (opens in a new tab)
- How long do you store my organization's data? Commercial plans (opens in a new tab)
- How do I change my model improvement privacy settings? (opens in a new tab)
- Custom data retention controls for Enterprise plans (opens in a new tab)
- Where are your servers located? Commercial products (opens in a new tab)
- Updates to Consumer Terms and Privacy Policy (signup choice, 5-year retention) (opens in a new tab)
- Use connectors to extend Claude's capabilities (opens in a new tab)
- Get started with custom connectors using remote MCP (opens in a new tab)
- Anthropic Trust Center (attestations) (opens in a new tab)
Your brief · 1 of 3
What brings you here?
See what we’d build first at a firm like yours.
Start with one process.
A working session on how your firm actually runs. You leave knowing the first process worth fixing.