Skip to main content
This page defines the operating model for administrators using Cogno on organization-managed Linux or macOS development servers. For the member connection flow, see Connect to a remote server.

Operating model

Members provide an SSH destination in Cogno Desktop. Cogno Desktop installs the Cogno CLI and the Engine service on that destination, registers the service with the OS, and connects to it over an SSH tunnel. Members and administrators do not run cogno server install in advance. The Engine service on the server runs the coding agent, holds the working copies and sessions, and reports to the Cogno API with the machine’s credential. Cogno Desktop reaches it only through the SSH tunnel — the Engine accepts no inbound network connections of its own. A machine is one OS account on one host. When several people use one server, each person sets it up from Cogno Desktop with their own OS account and SSH destination, and each appears as its own machine. Native Windows is not supported as a Remote Server. WSL is treated as a Linux environment.

Set up and update

1

Open Machines

The member opens Settings → Machines in Cogno Desktop.
2

Set up the server

Under Set up a new server, they enter the SSH destination and choose the coding engine. Cogno Desktop installs the Cogno CLI and the Engine service, issues the machine credential, and connects. Installing the coding engine is a separate, non-blocking step: if it fails, the server still registers and connects, and the failure is shown as a next step.
3

Choose the Project folder

For each Project, the member selects the repository folder in Project settings.
Updates are delivered from Cogno Desktop over the same authorized SSH session: the app verifies the release against its pinned trust root, pushes the binaries to the server, and runs cogno server update there. The server does not fetch binaries for itself, so a machine behind a proxy or with no outbound route still gets updated.
Servers set up before the Engine migration ran an earlier background service called WOD. The Engine retires it automatically: on startup it unregisters the old service from the OS and removes its state. No administrator action is needed. Current releases still install a wod helper binary next to the Engine in ~/.cogno/bin — it is not a service; coding agents invoke it as the local Cogno MCP command where a run uses the stdio server (see Manage Cogno MCP).

Cogno MCP

Coding agents reach Cogno’s task tools through Cogno MCP. For Engine sessions, the Engine passes the agent an inline MCP configuration for each run — a single HTTP server named cogno, authorized by a token scoped to that turn. Nothing is written to the server’s MCP configuration files, and the server itself exposes no MCP endpoint or inbound port. If the organization restricts which MCP servers a coding agent may use, the policy must allow the cogno server (tools named mcp__cogno__*) — and, where policy pins commands, the local stdio command ~/.cogno/bin/wod mcp cogno. See Manage Cogno MCP for configuration examples and validation.

Stop using a server

The member removes the machine under Settings → Machines → Your machines with Remove. Its credential is revoked, and the server stops running work as soon as its next check-in is refused.

Validation

  • Cogno Desktop can set up the SSH destination
  • The server’s row shows Tunnel connected in Your machines
  • A Project can select a folder on the Remote Server as its execution location
  • Work runs with the selected coding agent
  • When MCP is restricted, policy does not block mcp__cogno__* tools
  • After Remove, the machine cannot run work