An OpenAI coding agent that uses focused skills for repeatable workflows and MCP servers for live data, authentication, and controlled actions.
OpenAI distinguishes a skill, which teaches Codex or ChatGPT a repeatable workflow, from an MCP server, which provides live data and controlled actions.
That division is well suited to pages that help developers decide whether to package instructions as a skill, expose functionality through MCP, or combine both in a plugin.
Capability map
Each capability has a distinct user job, setup path, governance note, and first-party source. We do not publish unsupported combinations.
Mcp Servers
For Codex, an MCP server supplies live data, authentication, authorization, and controlled actions, while a skill teaches the repeatable workflow for using those tools. This division supports high-quality developer integrations because each page can explain the service boundary, required permissions, action sequence, and what the agent must not infer.
View setup and governance guidanceSkills
A Codex skill teaches a focused, repeatable workflow: tool sequence, decision points, output requirements, examples, and boundaries. The most useful implementation starts from a recognisable use case, defines the expected input and output, and separates reusable instructions from live systems that need MCP tools.
View setup and governance guidanceThese records support the agent-level claims on this page. Capability pages link to their relevant source separately.
Checked Aug 16, 2026
Checked Aug 16, 2026
Codex has 2 verified extension areas in this directory: Mcp Servers, Skills.
This page is sourced from 2 first-party documentation records, most recently checked August 16, 2026. Capability pages are published only when official support, setup guidance, and a source are present.