Skip to content

AI and MCP

In zAppDev, AI helps you describe your business application, and the engine produces the code. AI agents and people work on the same thing: the Intended Design. An agent does not guess your architecture from files. It reads the Intended Design, proposes a change to it, and the change is validated before any code is produced.

Why the Intended Design matters to an AI

Memory files, skills and specifications help an agent find its way, but they are advice: they cover what someone wrote down, they can go stale, and nothing checks the result. The Intended Design is different in kind:

  • It describes the whole application, across every kind of model, in one connected description.
  • It is checked mechanically by the engine, not by convention.
  • The code is produced from it, so there is one source to keep in line.

The checks cover structure, not meaning

The checks find broken relationships and wrong types. They do not find a business rule that is simply wrong, so tests still matter. See Unit tests.

Two ways in

zAI Your own agent
Where it runs Inside Studio In your own tool, such as Claude Desktop, Claude Code, Cursor, VS Code or Codex CLI
How it connects Built in Through the application's MCP server, with an access token
What it works on The Intended Design The same Intended Design
What checks it The same validation as a person The same validation as a person
  • zAI: the assistant inside Studio.
  • The MCP server: connect your own agent, and create and revoke access tokens.
  • MCP tools: what an agent can read, change, check and build.

AI agents connect through the MCP server that every application has, to read, change, validate and build the same Intended Design that people edit in Studio AI agents connect through the MCP server that every application has, to read, change, validate and build the same Intended Design that people edit in Studio

The loop

  1. An agent reads the part of the Intended Design it needs.
  2. It proposes a change, for example adding an association or an attribute.
  3. The Intended Design is validated. If the change cannot be supported, validation names exactly what is wrong, and the agent corrects it.
  4. Once the Intended Design is valid, the application can be built, and the engine produces the code.

In this section

  • zAI


    The assistant inside Studio. It reads the Intended Design through tools and proposes changes.

  • MCP server


    Connect your own agent to an application, with a scoped access token.

  • MCP tools


    The tools an agent can use to read, change, validate and build the Intended Design.

Where to start

  1. Try zAI in Studio. See zAI.
  2. Connect your own agent to an application. See MCP server.
  3. Know what validation can and cannot prove. See Behaviour to know.
  • Targets: what the engine produces once the Intended Design is valid.
  • Glossary: one term per concept.