Design Microservices with Codex and MCP

Install the graph-aware authoring tools

Follow plugin installation, then open a new Codex task in the typed Python topology project. The project manifest identifies the architecture entrypoint, canonical output and generation targets.

MCP resources expose model context; tools perform explicit inspect, validate, export and generation operations. The packaged workflow is a Codex integration, not a claim that a separate ChatGPT connector is installed.

Ask for a change with explicit boundaries

State the intended behavior, service ownership and failure path. For example, ask for a Kafka event boundary after order processing while preserving the existing HTTP response contract. Ask the agent to explain the architectural diff before generating the project.

Inspect the current order topology. Add an outbound Kafka event contract
and a separate Analytics consumer. Preserve the existing HTTP response
contract. Show the changed services, endpoints, types and links, then
validate the model before exporting it.

Review semantics, not only syntax

Check that cross-service communication uses endpoints, independent work has the intended call semantics and errors have an explicit route. Validation can reject invalid topology, but does not prove that the business intent is correct.

Resolve the authoritative validator diagnostics, export the model and generate an archive only after the change is understood. Preserve user-owned business files through safe regeneration.

Next steps

Inspect → change typed declarations → review architectural diff → validate → export → generate

Related documentation