Codex integration guide

AI service design with Codex and MCP

The Service Architect plugin lets Codex inspect, author, validate, export, and generate projects through a graph-aware interface.

When to use

Use the plugin when an agent should change architectural concepts directly while deterministic code owns serialization and validation.

Codex edits the typed Python project: services, pipelines, streams, connectors, endpoints, types, pools, modules, and links remain explicit objects.
MCP tools expose bounded operations such as inspection, validation, YAML export, import, and runtime project generation.
Passive MCP resources expose workspace state, ServiceGen capabilities, and versioned validation metadata without executing authoring code.
A semantic authoring playbook tells the agent how architectural concepts combine, not merely which Python methods exist.
You can use ChatGPT to discuss and plan service architecture, while the packaged tool integration documented here runs in Codex through MCP.

IntentDescribe the topology changeUse architectural language: services, streams, transports, pools, retries, and schedules.

AuthorModify typed PythonConcrete factories expose only valid properties and call semantics.

ValidateResolve SG_* diagnosticsRead the exact versioned rule and remediation supplied by ServiceGen.

DeliverExport or generateProduce canonical YAML or download a native runtime project ZIP.

The plugin connects existing product layers instead of introducing another model.

PropertyTypeDescription

sa-python-dslAuthoring API

Typed editable topology and deterministic YAML serialization.

Service Architect pluginCodex plugin

Skills, MCP server configuration, and semantic authoring instructions.

Codex MCPAI architecture integration

Turns natural-language service design intent into bounded typed-model operations.

ChatGPTPlanning and discussion

Useful for architecture discussion; direct repository tools in this guide are provided by the Codex plugin.

ServiceGenAuthority

Validation diagnostics, capability registry, and generated runtime projects.

DesignerVisual interchange

Loads exported YAML and reconciles visual graph changes with the Python project explicitly.

User intent
  -> Codex + semantic playbook
  -> typed Python project
  -> ServiceGen validation contract
  -> canonical YAML or generated project ZIP