Skip to main content
RelayHub can connect to outside AI assistants like Claude and ChatGPT so they can work with your RelayHub data on your behalf. An assistant you connect can search your knowledge, run your skills and agents, and work with your files, all within the same permissions you already have. This uses an open standard called the Model Context Protocol (MCP), so any client that speaks MCP can connect.
This page is about an outside assistant connecting IN to RelayHub. If instead you want RelayHub to reach OUT and use an external tool server’s tools inside your chats and agents, see Connected Services.
You are always in control. Nothing connects without your approval, a connected assistant only ever sees what you could see yourself, and you can cut off any connection at any time.

​
How a connection works

When you set up an MCP client (for example Claude Desktop), you point it at your RelayHub connection address, which is your RelayHub web address followed by /mcp. For example, if you sign in at https://acme.relayhub.ai, the connection address is https://acme.relayhub.ai/mcp. The first time the assistant connects, RelayHub sends you to a sign-in and approval page. You log in the same way you always do, then review what the assistant is asking for and approve it. After that the assistant can work with your RelayHub data until you revoke it.
A connection acts as you. It can only reach the workspaces, files, skills, and agents that your own account can reach. It cannot see another person’s private data, and it cannot do anything your account is not allowed to do.

​
What a connected assistant can do

When you approve a connection, you approve it for one or more of these areas:
  • Read: list and open your agents, skills, workspaces, and conversations, and search your knowledge and knowledge graph.
  • Run: run a skill or an agent, send a chat message, or start an automation.
  • Files: list your files, open a file, or add a new file to a workspace.
The approval screen shows exactly which areas the assistant is requesting before you agree. You can approve a read-only connection if you only want the assistant to look things up, without letting it run anything or change files.

​
Reviewing and revoking connections

You can see every assistant you have connected, and remove any of them, under Settings, in Authorized Connections. Each entry shows the assistant, when you connected it, and what it can do. Removing a connection takes effect immediately: the assistant loses access on its very next request, with no waiting period. If you ever connect an assistant by mistake, or you are finished using one, remove it here.
Every action a connected assistant takes is recorded in your audit history, the same as if you had done it in RelayHub yourself. Administrators can review this activity in the Audit Hub.

​
Good to know

  • You can connect more than one assistant, and each one has its own separate approval and its own entry in Authorized Connections.
  • A connection is tied to your account. It does not give the assistant a way to sign in as anyone else.
  • If your administrator has turned on strict security policies for your organization, connections still follow those same rules, including which AI providers and models are allowed.