Connectors

A connector is a registered MCP server that you use as a tool in conversations and agents, from the SDKs or the API. You don't run an MCP client yourself.

After you register a connector, the model discovers its tools and calls the ones that match the request of the user. You can also call a tool directly, without the model.

The Model Context Protocol (MCP) is an open standard that connects AI models to external tools and data sources. You can register any MCP-compatible server as a connector.

API keys have a Connector access scope. The default scope, Shared connectors only, gives access to the connectors shared with the workspace, not to your private connectors. To use your private connectors with a key, create the key with the Private and shared connectors scope. For details, see API keys.

• Quickstart: register a public MCP server, connect it, and call one of its tools.
• Manage connectors: register a connector with the MCP server URL and a visibility scope.
• Debug connectors: test an MCP server from Studio before you use it.
• Use connectors in conversations: pass a connector in the tools array and let the model pick the tool.
• Call tools directly: call a specific tool when you know which one and with which arguments.
• Human-in-the-loop: stop tool calls until a user approves them.