Verifying an installation

The same three checks verify any installation, on Claude Desktop chat, Claude Cowork and the Code tab of the Claude desktop app, and in VS Code. Run them in order and stop at the first failure. They are also the preflight ea-start-here runs at the start of every session. You ask the assistant to run them. The calls are deterministic, and the assistant chooses to run them when you ask.

Start Enterprise Architect and open any project before the second check.

1. ping. It does not touch EA. The ea-start-here skill calls it with the release of the skills the session loaded, and it reports that the server is alive, which version it is, and whether the server or the skills are behind the current release. When either is behind, the reply carries a notice with the steps to take, and the skill shows it word for word. A reply with no notice means both are current. If nothing answers, the AI Power Tools MCP Server for Sparx EA is not running: the Claude Desktop extension is not installed or is switched off (Claude Desktop extension), and nothing else will work. Skills alone do not bring the server. See Installation troubleshooting.

2. get_repository_info (in ea_repository). This is the first check that reaches EA. A success reports the open project's path and connection, the EA version, the repository type, whether user security is enabled and who is logged in when it is, and the top-level packages. When a newer server release exists it also says so. When skill copies left by the older skills installer remain, it lists them (Moving from an older installation). Failures are structured:

   Error	Meaning	Remedy
   ea_not_running	EA is not running on this machine	Start EA yourself, or let the assistant call launch_ea with your agreement. Where computer use is available, the assistant can instead start EA through its window, after asking you
   ea_no_model_open	EA is running but no model is open	Open a model with open_model, or create a new one with create_model. Where computer use is available, the assistant can also open it through EA's window

3. assess_mdg_situation (in ea_mdg). It reports which modeling technologies are loaded, how much of the model conforms to them, and how many stereotypes are used outside any technology. It confirms that the ea_mdg tool and the model-reading path work end to end.

What a good result looks like

• ping answers with the server version you expect and no update notice. A notice is not a failure: it names what is behind and the steps to bring it current (Updating and uninstalling).
• get_repository_info names the project you expect. Writing to the wrong repository is the most expensive mistake available and it is silent, so read the project path.
• assess_mdg_situation returns without an error.
• The skills are visible: ask for ea-modeling by name. Each skill appears once. A skill that appears twice means a second copy is loaded (Installation troubleshooting).
• In a local Claude Cowork session the EA tools respond. A cloud session has no local MCP server, so no AI Power Tools tool is present.

In VS Code, the server comes from the extension and the skills from the ai-power-tools plugin installed in VS Code (VS Code extension). VS Code's Claude chat mode also loads the plugin installed for Claude, so on a machine with both it lists each skill twice.