zAppDev Create free account

AI · zAI and any MCP agent

Every AI change, checked before it becomes code

zAI and any MCP agent change the Intended Design of your application, and every change is checked before any code is produced. The finished application is agent-ready too: every service can become an MCP tool.

Create free accountSee the difference
✓Checked before code is produced✓Open to any MCP agent✓Agent-ready applications

Element types and their relationships

governsreadsreads · writescallsrunsbinds tostylesmaps structs of zAIor any agent Intent BusinessObject Thelanguage APIAdapter Theme Form DataMapping governsreadsreads · writescallsrunsbinds tostylesmapsstructs ofzAIor any agentIntentThelanguageBusinessObjectAPIAdapterFormDataMappingTheme

Every relationship is typed.

The difference, in one picture

An AI can assemble your architecture. Or you can give it, checked.

Good agents already use memory, skills and instruction files to find their way. Same request in both panels: add an approval step.

Today, with the best tooling

Context is assembled

Reading files, notes and instructions→Combining what it found→Proposing a change→Done. Does it hold?
1
Codeit can read only what fits in view
UserService.csOrderController.csapproval.tsOrderForm.html
2
Adviceit was given, written by people, can go stale
memoryorder rulesskillcode stylespecSPEC.md
3
What it cannot seenever seen, or never written down
pricing.tsaudit-log.csdb-migration.sqllimits.jsonrole-check.cs

A rule nobody wrote down is broken, and nothing checks it

vs
With zAppDev

Context is given

Reading the whole architecture: every view and relationship at once→Proposing a change→Checked. Producing the code
1
The whole architectureevery view, in one Intended Design
1 Intent2 Domain3 Logic4 UX/UI5 Integration6 Data
2
Checked mechanicallynot advice: the checks
Relationships validTypes matchPermissions hold
3
Code produced from itregenerated, so it stays in line
for your target stackOrdinary code, yours

The change is checked against the whole Intended Design

Memory, skills and instructions help. They carry what someone remembered to write down, and they are advice: an LLM can ignore them, and they go stale. A checked Intended Design is different in kind. It describes the whole system, it is checked mechanically, and the code is produced from it, so there is one source to keep in line. Structure is checked; behaviour still needs tests.

zAI at work

It reads the application, before it answers

This is zAI in Studio, asked to summarise the operations of a business object. Look at how it goes about it.

Studio · zAI Assistant at work
zAI reading the application in Studio
1

It reads the Intended Design, through tools

zAI calls the same tools any MCP agent can use, such as listing the classes of a business object and getting their details.

2

It looks at the real definitions

It retrieves the class and operation definitions, not a guess about what they probably contain.

3

It reads the actual code

To summarise an operation, it inspects the code itself, rather than inferring from names.

A real session in Studio on the Web application template. zAI-generated content may not be accurate, so changes are reviewed.

Open to any agent

Every application comes with its own MCP server

You do not have to use zAI. Whichever agent your team prefers can read, change, validate and build the same checked Intended Design, under the same checks.

Your agent

Any MCP client

→
One per application

MCP server

→
Checked

Intended Design

→
Produces

The engine

Claude DesktopClaude CodeCursorVS CodeCodex CLIAny MCP client
Studio · MCP access tokens
MCP access tokens in Studio
1

An MCP server for every application

Each application has its own MCP server, so any agent can work on it with the same tools that zAI uses.

2

Connect the agent you already use

Create an access token, then connect clients such as Claude Desktop, Claude Code, Cursor, VS Code or Codex CLI, or any other MCP client.

3

Scoped, and revocable

Each token is scoped to one application, can carry an expiry date, and can be revoked at any time.

Agent names are trademarks of their owners and are listed as clients that can connect.

The self-correction loop

Wrong answers fail early and clearly

When the AI proposes something the graph cannot support, validation says exactly what is wrong. The AI reads the message and fixes it, before anything is built. Structural mistakes are caught before the build, not in a failed release.

AIAdd Order.Customer
The checksUnknown element “Custmer”
AIFix: Order.Customer to “Customer”
The checksIntended Design is valid
The engineApplication produced

Illustrative validation loop.

Agent-ready output

Your services become tools agents can call

A service of type MCP exposes each of its operations as an MCP tool, registered automatically in the application’s MCP server. An agent can call your own business logic, not only edit the Intended Design.

Studio · Service editor, type MCP
The service editor with type MCP
1

Choose the type

A service is either REST or MCP. Choose MCP for the tools you want agents to call.

2

Operations become tools

Each operation you write, in the language, is exposed as a tool call.

3

Registered automatically

No wiring: the tools appear in the application’s own MCP server, which is produced for you.

Give your AI something to reason about

Create a free account and let zAI work on a real Intended Design of your application.

Create free account Read the documentation