MCP Servers

MCP servers let you expose your workspace tools to external AI clients like Claude Desktop, ChatGPT, and Cursor. Clients connect securely using the OAuth 2.1 protocol, so each connection is authorized and scoped to a specific server.

Who can create MCP servers

  • Any workspace member with access to tools can create an MCP server.
  • You must have at least one tool available in your workspace before you can create a server.

Who can connect to an MCP server

  • Only the workspace member who created the server can authorize client connections.
  • Each client must go through the OAuth authorization flow before it can access the server.

What connected clients can do

  • Call any tool that was included in the MCP server at creation time.
  • Send multiple requests using the same authorization until the token expires or is revoked.
  • Automatically refresh expired tokens without requiring you to re-authorize.

What connected clients cannot do

  • Access tools that were not included in the MCP server.
  • Access other MCP servers in your workspace.
  • Perform actions outside the granted scope.
  • Continue accessing tools after you revoke the server.

How MCP servers work

When you create an MCP server, you choose a name and select which tools to include. The server gets a unique URL that you provide to your AI client. When the client connects for the first time, it redirects you to an authorization page where you can review the requested tools and approve or deny access. Once approved, the client receives a time-limited token to make requests on your behalf. Tokens expire after one hour but are automatically refreshed by the client, so you typically only need to authorize once.

If you revoke an MCP server, all active connections are immediately terminated and the server URL stops working.

Supported clients

MCP servers work with any MCP-compatible application. Here are setup guides for popular clients:

Related guides