Vai al contenuto

MCP Servers

Model Context Protocol (MCP) is an open standard that lets AI models use external tools and data sources through a common interface. SkyDeck.ai can connect to MCP servers so that the models in your workspace can call those tools during a conversation.

There are two kinds of MCP servers:

  • Remote servers are hosted in the cloud and managed by SkyDeck.ai. An administrator adds a server to the workspace and supplies any configuration it needs — for example, an API key the server uses to reach a third-party service.
  • Local servers run on your own computer through the SkyDeck.ai desktop app and are invoked from your conversation. Because they run locally, they can work with resources on your machine. Each member has their own local server.

Setting up MCP servers (Control Center)

Workspace administrators manage MCP servers in the Control Center. For a workspace MCP server, an administrator can:

  • Provide environment variables — many servers need credentials or settings (such as an API token) in order to run. These are stored securely and passed to the server when it runs.
  • Control access with tags — assign tags to a server so that only members with matching tags can use it, the same way access is managed for other tools.
  • Enable or disable individual tools — a single server can expose several tools, and an administrator can turn specific tools on or off for the workspace.

SkyDeck.ai discovers the tools that each server provides, so the available tools appear automatically once a server is connected.

Using MCP tools (GenStudio)

In GenStudio, members can use the MCP tools they have been given access to. A member can also further restrict which MCP servers and tools are active for their own account — a member's settings can only narrow the access an administrator has granted, never widen it.

Note

MCP support is a newer capability and may need to be enabled for your workspace. If you don't see MCP options in the Control Center or GenStudio, check with your workspace administrator.

For more about the protocol itself, see the Model Context Protocol documentation.