Skip to content

Connecting an AI Assistant to QIE

The Model Context Protocol (MCP) is a standard way for AI assistants, such as Claude Code or Codex, to connect to other software. QIE hosts an MCP server, so people can ask an AI assistant to work in QIE for them: build, change and test channels, find out why messages errored, and look through messages and logs. The assistant signs in as the person using it and can do only what that person is allowed to do. Every call is recorded in the MCP audit log.

Connecting an assistant takes two steps: an administrator turns the MCP server on once, and then each person adds the server to their own AI assistant and approves it.

Turning the MCP Server On

An administrator does this once for the whole installation. Every node in a high availability cluster uses the same settings.

  1. Make sure Global Settings holds a certificate, with its private key, for the host name clients will use. Use one issued by your organization's certificate authority.
  2. Open System Administration -> Model Context Protocol (MCP).
  3. Tick Enable the MCP server, enter an unused MCP Port (View Ports lists the ports QIE already uses), and choose the Server Certificate.
  4. Check the Resource URL. It is the address AI assistants use to reach the MCP server, ending in /mcp, and by default it is https://<server name>:<MCP port>/mcp. It must use a host name every person's computer can resolve, and the certificate must be issued for that name.
  5. Enter an Environment Label, such as PROD or TEST, so people can tell two engines apart.
  6. Click Save.

Then open the MCP port to the workstations that will connect, and only to those. Each setting is described in MCP Settings, and the page itself in Model Context Protocol (MCP). Read MCP Server Security Guidance before turning the server on in production.

Connecting Your AI Assistant

Each person connects their own AI assistant, on their own computer.

  1. Ask an administrator for the Resource URL.
  2. Add the server to your AI assistant with that address, as the page for your assistant describes: Connecting Claude Code or Connecting Codex.
  3. Sign in. Some assistants start the sign-in when they first connect, and with others you start it yourself, as the page for your assistant says. The assistant opens the QIE console's approval page in your browser: sign in to the console if you are not signed in, choose the access to grant and click Approve. Read only is the default; grant manage or full access only when the work needs it.
  4. Ask your assistant to work in QIE. The build a channel prompt starts it building a channel from a description of what the channel should do.

An assistant can connect only from the person's own computer. Assistants that run in a vendor's cloud cannot sign in (see Supported Clients).

What the Assistant Can See

Return message content on the Model Context Protocol (MCP) page is off by default. While it is off, the assistant sees channels and other configuration, statuses and message metadata, but no message bodies, log text, cache values or sample messages. See Message Content and the MCP.

Actions that cannot be undone, such as deleting, purging or importing, ask you to confirm in your assistant before they run. An assistant that cannot ask is refused them while Require client confirmation is on, which is the default (see MCP Tool Results and Errors).

Working Through the Remote Management Hub

An organization that manages its engines from a Remote Management Hub can sign AI assistants in once, at the Hub, instead of at each engine. The Hub runs its own MCP server and sends each tool call to one site, an engine registered with it, with a statement it signs for that call. A site runs such calls only while it accepts MCP calls through the Hub.

More Information