OpenAI Plugins Documentation — Skills, MCP Servers and UI for ChatGPT and Codex
by OpenAI
OpenAI's first-party spec for building, packaging and getting a ChatGPT plugin through review.
Overview
The site is organised into eight sections that follow the lifecycle of a plugin. Plan covers brainstorming use cases and defining tools. Concepts explains the three building blocks: the plugin package itself, Skills (a self-contained folder holding a SKILL.md plus supporting files that give the model repeatable workflow instructions, used when the model's existing tools already suffice), and an MCP server that exposes callable tools with typed input and output schemas, along with UI guidelines that argue custom UI is optional and only justified when people need to inspect, compare, edit, confirm or navigate structured information. Build is the hands-on core: 'Build an MCP server' walks through choosing the TypeScript or Python MCP SDK, designing focused tools, returning structured results with no custom interface, adding authentication, testing locally with MCP Inspector and deploying to a production HTTPS endpoint; sibling pages cover Build skills, Add UI to your MCP server, the Checkout API for monetisation, packaging and an MCP-server-plus-UI quickstart. Deploy documents connecting and testing through ChatGPT developer mode, the MCP server review requirements, submission, submission errors and troubleshooting. Guides add metadata optimisation, security and privacy, three commerce conversion specs (product checkout, restaurant reservation, local-services quote) and a page on submitting a Claude Code plugin to OpenAI. Every page is published as Markdown with a machine-readable index at /plugins/llms.txt.
At a Glance
- Topic
- MCP
- Level
- Intermediate
- Format
- Documentation
- Cost
- Free
- Duration
- ~1-2 hours for the quickstart and a first MCP server; ~6 hours to read end to end
- Provider
- OpenAI
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Build an MCP server in TypeScript or Python that ChatGPT and Codex can call
- ✓Decide whether a plugin needs a Skill, an MCP server, or both together
- ✓Write SKILL.md folders that give the model repeatable workflow instructions
- ✓Design focused tools with typed input and output schemas the model picks correctly
- ✓Add optional React UI components to an MCP server's tool responses
- ✓Authenticate users with OAuth and stand up a production HTTPS endpoint
- ✓Test locally with MCP Inspector and ChatGPT developer mode before submitting
- ✓Meet the published MCP server review requirements and debug submission errors
- ✓Optimise plugin metadata so the model invokes your tool at the right moment
Highlights
- •The published review requirements and submission-error list turn 'will OpenAI accept this?' from folklore into a checklist
- •Built directly on the Model Context Protocol, so an MCP server you already run is most of the work done
- •Covers the commercial path unusually thoroughly: a Checkout API plus conversion specs for checkout, reservations and local-services quotes
- •Companion repo openai/openai-apps-sdk-examples (2.3k stars, MIT) ships runnable widgets — carousel, map, shopping cart, 3D solar system — with both TypeScript and Python MCP servers
- •A dedicated guide for porting an existing Claude Code plugin to OpenAI's surfaces
- •The whole site is served as Markdown behind an llms.txt index, so you can point a coding agent at it directly
Who It’s For
Best For
- ✓Engineers who already run an MCP server and now want distribution inside ChatGPT
- ✓Product teams planning a ChatGPT-native app or an in-chat commerce flow
- ✓Developers porting a Claude Code plugin or Agent Skill to OpenAI's surfaces
- ✓MCP developers who need to pass a formal review process rather than self-host
Prerequisites
- •Working TypeScript or Python and comfort running and deploying a web server
- •Familiarity with the Model Context Protocol — tools, schemas and transports
- •A ChatGPT account with Developer mode enabled, which the quickstart requires
FAQ
What is OpenAI Plugins Documentation — Skills, MCP Servers and UI for ChatGPT and Codex?
The official OpenAI documentation for building plugins that run inside ChatGPT and Codex, written for engineers who want distribution rather than a local integration. It covers the three building blocks — Skills, MCP servers and optional UI — plus authentication, packaging, the review requirements and the submission flow, so you finish able to ship a plugin OpenAI will actually accept.
Is OpenAI Plugins Documentation — Skills, MCP Servers and UI for ChatGPT and Codex free?
OpenAI Plugins Documentation — Skills, MCP Servers and UI for ChatGPT and Codex is free to access.
What level is OpenAI Plugins Documentation — Skills, MCP Servers and UI for ChatGPT and Codex for?
OpenAI Plugins Documentation — Skills, MCP Servers and UI for ChatGPT and Codex is aimed at a intermediate audience. Recommended background: Working TypeScript or Python and comfort running and deploying a web server, Familiarity with the Model Context Protocol — tools, schemas and transports, A ChatGPT account with Developer mode enabled, which the quickstart requires.
How long does OpenAI Plugins Documentation — Skills, MCP Servers and UI for ChatGPT and Codex take?
Expect roughly ~1-2 hours for the quickstart and a first MCP server; ~6 hours to read end to end. Most learners work through it at their own pace.
What will I learn from OpenAI Plugins Documentation — Skills, MCP Servers and UI for ChatGPT and Codex?
You'll learn: Build an MCP server in TypeScript or Python that ChatGPT and Codex can call; Decide whether a plugin needs a Skill, an MCP server, or both together; Write SKILL.md folders that give the model repeatable workflow instructions; Design focused tools with typed input and output schemas the model picks correctly; Add optional React UI components to an MCP server's tool responses; Authenticate users with OAuth and stand up a production HTTPS endpoint; Test locally with MCP Inspector and ChatGPT developer mode before submitting; Meet the published MCP server review requirements and debug submission errors; Optimise plugin metadata so the model invokes your tool at the right moment.
Topics
Sources
This page was written from 4 sources, 1 on domains other than developers.openai.com.