Codex app-server is the component through which OpenAI exposes the Codex agent harness — the system that powers the Codex app, CLI and IDE extension of OpenAI Codex — to other applications. According to Codex as a platform: build on the open agent harness, it offers the harness’s capabilities through a documented client protocol: applications can create threads, start turns, receive events and handle approval requests. OpenAI publishes it as open source, alongside the Codex CLI and the official Codex SDK.
OpenAI’s pitch for it is that a team building software that needs an agent can start with Codex instead of inventing a new runtime, and then decide what the surrounding application should own.
Capabilities
The post describes app-server as letting an application connect to a local Codex process, keep conversations open, stream events, interrupt work, expose tools and respond to approval requests. Underneath, the Codex harness manages conversation state, streams execution, uses tools, enforces configured sandbox and approval policies, and carries work across turns.
OpenAI positions it as one of three ways to build on Codex: codex exec for bounded non-interactive jobs such as a script or CI task, the Codex SDK for programmatic workflows that start, resume or stream tasks, and app-server “when the agent is part of the product itself”. In the post’s words, the SDK simplifies common programmatic workflows, while app-server gives product teams direct control over the lifecycle and user experience.
Adoption & Ecosystem
The division of labour OpenAI describes is that the application owns product context, business rules, tools and consent, while app-server provides the agent loop and sandboxed execution. Its worked example is Relay, a sample operations application OpenAI built on app-server: an agent sits beside a fictional shipment dashboard, connected to application-owned MCP tools (Model Context Protocol), and must obtain human approval before rebooking a shipment. The harness handles the agent loop, conversation state, streamed activity and tool interaction, while the product keeps its dashboard, records and controls.