Orion

Getting started ​

Orion hasn't had its first release, so for now you build it from source. CONTRIBUTING.md covers the toolchain and every build.

The desktop app ​

The desktop app bundles the canvas with the orion CLI and runs its server for you. Build it with npm run build in workspace/orion-desktop/, then move the app it produces into your Applications folder.

The packaged app keeps one local address across restarts, so each canvas remembers its viewport and detail level.

The CLI ​

The CLI reads and changes the same data as the app, and the app picks up every change while it's open. The desktop app ships the binary at /Applications/Orion.app/Contents/MacOS/orion, and a local build writes workspace/orion-cli/build/orion. Put one on your PATH, then run:

sh
orion help

See the CLI reference for every command, or run orion help --all to print them all in the terminal.

Connect an agent over MCP ​

orion mcp start runs an MCP server over stdin and stdout, so AI clients can discover Orion's canvas tools and call them. Each tool runs an existing CLI command in the same process, so the CLI stays the only writer and never runs a shell command built from tool input.

Add Orion to an MCP client as a stdio server that runs orion mcp start, using the full path to the binary. Most clients accept a config entry like this one, or an equivalent mcp add command:

json
{
  "mcpServers": {
    "orion": {
      "command": "/Applications/Orion.app/Contents/MacOS/orion",
      "args": ["mcp", "start"]
    }
  }
}

Pass --canvas <name> to keep the server on one canvas. Otherwise each call uses the selected canvas unless the tool call names one. Set ORION_DATA_DIR in the server's environment to use another data folder.

See the MCP reference for every tool.

Next steps ​