AI Power Tools for Sparx EA Docs Release 3.6.4

How it works

Layers

From top to bottom:

Layer What it is Where it runs
Client Claude Code in the Claude desktop app, Claude Cowork, Claude Desktop chat, or VS Code with GitHub Copilot in Agent mode Your workstation. A local Claude Cowork session runs on the workstation. A cloud session has no local MCP server and cannot reach EA
Plugin and skills The ai-power-tools plugin: the skills the client loads. It carries no server Loaded by the client
Server ea-mcp-server.exe, a signed Windows program the client starts and talks to. Claude starts it from the Claude Desktop extension The same Windows machine as EA
EA automation interface EA's own Windows automation object, reached through Windows' component system (COM) The same machine
Enterprise Architect The running EA application with a project open The same machine
Repository A local file (.qea or .qeax), a server connection to a shared repository held in a database, or a cloud connection to a shared repository through Pro Cloud Server Wherever EA's connection points

The server must run in the same Windows environment as EA, because EA's automation interface does not cross machine boundaries here.

The four capabilities

The layers above are the path for the server. The server is one of four capabilities the assistant works with, and the skills use whichever of them your client provides.

Capability What it gives the assistant
MCP server The 149 EA operations, including the reporting build, its checks and its refresh. Required for all EA work
Computer use Seeing and operating desktop applications: EA's own window, and Power BI Desktop. It reaches EA directly, not through the server
Local code execution Running the code engines bundled with the skills, on the machine where EA runs: for example the diagram engines, the census and governance gate that prepare a reporting build, and the Power BI project generator. They give the same result every time and use far fewer tokens than the AI working the same thing out
Browser control Operating web applications, through a built-in browser or Chrome. All Prolaborate work uses it

Which capabilities you have depends on the client you work in. The matrix by client, what each capability enables, what happens without it and how to check your own session are in Capabilities and surfaces.

How a request flows

  1. You You ask in plain language: "Create a Business Application element named Core Banking in the Applications package."
  2. AI The client picks a skill if one matches, reads its guidance, and decides which operations to call and in what order.
  3. AI The client calls a tool. Every operation goes through one of the six grouped tools, which are each named for an area of work (ea_model for model content, ea_diagram for diagrams, and so on). The call names the operation and carries its arguments.
  4. AI Power Tools The server checks the call before touching EA. The operation name must be well formed and known. Read-only mode must not block it. Every argument must be one the operation accepts, and every required argument must be present. A call that fails a check returns a structured error the assistant can read and correct.
  5. AI Power Tools The server calls EA. It attaches to the running EA instance on first use and reuses the connection. If EA is not running, or no model is open, it says so and names the remedy. The server itself does not start EA. The assistant can: where computer use is available it starts EA and opens the model through EA's window, after asking you, and retries. Where it is not, it asks you to do it.
  6. AI Power Tools EA performs the work against the open repository. Some operations read or write the repository's database directly through EA. The reporting operations that build SQL views connect to the database themselves: to SQL Server with the Windows account that runs the server, or to a SQL database in Microsoft Fabric with a Microsoft Entra sign-in that you complete.
  7. AI Power Tools The server returns a result, marked as succeeded or failed with an error code, and for diagram operations often an inline image. Results are kept small by default.
  8. (AI + AI Power Tools checks) The assistant verifies and reports. Skills tell it to re-read or re-render rather than trust a success message, because EA can accept a write and not apply it.

This is the path of a server operation. Work done with the other capabilities takes a different path: the assistant operates an application's window (computer use), runs a code engine (local code execution) or drives a web page (browser control), and then checks the result as each skill directs.

Where each deliverable lives

Deliverable Contains Reaches the machine through
ai-power-tools plugin The skills only Your Claude account, from the novocircle marketplace. See Claude plugin
AI-Power-Tools-for-Sparx-EA.mcpb The server only The Claude Desktop extension pane. One install serves Claude Desktop chat, Claude Cowork and Claude Code in the Claude desktop app. See Claude Desktop extension
ai-power-tools-ea.vsix The server, registered with VS Code, and the setup of the skills plugin in VS Code VS Code. See VS Code extension