Skip to main content
The same names play two roles. As a harness, the client runs inside xpander in a cloud session. As a desktop client, it runs on a laptop and reaches xpander over MCP.

​
As harnesses inside xpander

Each conversation runs in its own sandbox with memory and CPU limits, gets the skills your organization made available and a short-lived AI gateway token instead of a provider key, and is recorded on its task. The composer can switch harness or model, and the history carries over.

​
As desktop clients

Any MCP client (Claude Code, Claude Desktop, Cursor, ChatGPT or another) can connect to either endpoint. The client keeps its own model sign-in, including a personal Claude plan (Pro or Max) through /login. The MCP connection signs in to xpander separately and never touches the model sign-in.

​
Which endpoint to connect

  • A person’s own client: Omni. One URL covers everything xpander Chat offers, and “Deploy this agent to xpander as a scheduled task” ends with an agent running in a cloud session. Setup: Connect your MCP client.
  • An integration or automation: the API endpoint. Raw calls, with no agent deciding in between. Setup: Model Context Protocol.

​
On a self-hosted install

On Air-Gapped, the endpoint exists once global.mcp.publicUrl is set. Until then, the MCP row on Platform > Agents > Channels reads Not offered on this install. On Hybrid, the chart serves mcp.<your domain> with nothing to set. If either page says the gateway is not available, contact xpander support. A desktop client can also reach the organization’s skills directly at the address on Platform > Agents > Skills registry, with each person signed in as themselves (MCP gateway).

​
What is different between the two roles

​
Next steps

Connect a client to Omni

Setup per client.

Connect a client to the API

The API over MCP.

Send laptop model calls through the gateway

Use an xpander key.