An assistant built into your application that knows the project the user has open and does the work they ask for — and a remote MCP server so Claude, Cursor, VS Code and other AI clients can drive the application the same way.
Your users open a chat panel and ask in plain language. The assistant is aware of the signed-in user and the open project, so “list the cables in this project”, “add two of these speakers and mount them in rack 1” or “patch every access point to the core switch” are answered in context. It does not only describe what to do: it reads the data, carries out the actions, checks the result and reports back, asking a clarifying question when a request is ambiguous.
The same tools are published through a remote Model Context Protocol server at your application’s address, with OAuth 2.1 sign-in and consent, so an external assistant works with your projects exactly as the built-in one does. A second endpoint offers administration tools. The server advertises itself with a standard server card, and lists only the tools the signed-in user is allowed and the modules the application has.
Every action the assistant takes goes through the application’s own API with the user’s own permissions: a user can never do through the assistant anything they could not do by hand. An administrator switches the assistant on or off for the whole application.