# The Codex and Claude plugin economy

I wanted to test a popular analogy: that plugins for Codex, ChatGPT and Claude will become the next app stores. I asked for a current census of what exists, what people use, what is missing and who captures the money. I told the tool not to assume the analogy was right.

## What I found
- Distribution is moving into the apps that host the plugins. OpenAI documents one shared plugin directory for ChatGPT and Codex. Anthropic put connectors, plugins, partner purchases and service partners under one Claude Marketplace.
- The two companies differ on money. OpenAI's plugin guidelines currently allow commerce through a plugin only for physical goods, with checkout on your own domain, and forbid selling digital products or subscriptions, so access through an existing subscription and selling a new digital subscription are different questions. This restriction alone does not establish where customers discover the product. Anthropic's marketplace lets customers buy partner products against committed spend.
- Counts are not comparable. Claude's plugin page said 528 plugins and its marketplace says over 2,000 connectors and plugins. These are different units, and neither tells you how many independent products exist. By 8 November 2026 the plugin page showed 529.
- Official listing pages show large install counts. Frontend Design shows 1,134,112 installs and Superpowers 1,009,371 as of 8 November 2026. The study could not establish what counts as an install, so these are not usage or revenue.
- Native tools already cover the obvious gaps. Claude documents plugin evaluation, cost and usage measurement, dependency constraints and organization controls. The study argues that a dashboard or test runner sold as missing infrastructure is a weak idea. It did not measure this.
- The money often sits behind the plugin. Context7 and Adspirer publish pricing for the service the plugin connects to, and the plugin brings customers to that service.
- Reliability emerged as a lead worth investigating. Public issue threads describe MCP connections that expire and stall automations. These threads could not be confirmed against a primary source. The study's own check found the cause narrower than "tokens never refresh", so it counts the threads as leads and does not size the market from them.

The study points toward a paid service that keeps a company's integrations working across different apps. I would test that before building a marketplace. I have not tested it with buyers.

## How it was made
I asked Perplexity Deep Research on 3 November 2026 and it used 70 sources. The AI gathered the sources and drafted the study. I asked the question. Figures were current at that date and may have changed.

## Sources
- [OpenAI: Plugins](https://learn.chatgpt.com/docs/plugins)
- [OpenAI: Package your plugin](https://developers.openai.com/plugins/build/plugins)
- [OpenAI: Plugin guidelines](https://developers.openai.com/plugins/plugin-guidelines.md)
- [OpenAI: Plugin submission](https://developers.openai.com/plugins/deploy/submission.md)
- [Claude Code: Plugins overview](https://code.claude.com/docs/en/plugins)
- [Claude Code: Plugin evaluation](https://code.claude.com/docs/en/plugin-evals.md)
- [Claude Code: Measuring plugins](https://code.claude.com/docs/en/plugins/measure.md)
- [Claude: Manage plugins for your organization](https://support.claude.com/en/articles/13837433-manage-plugins-for-your-organization)
- [Claude Marketplace](https://claude.com/marketplace)
- [Claude: plugins directory](https://claude.com/marketplace/plugins)
- [Claude: Frontend Design plugin](https://claude.com/marketplace/plugins/frontend-design)
- [Claude: Superpowers plugin](https://claude.com/marketplace/plugins/superpowers)
- [Claude Code: Plugin dependencies](https://code.claude.com/docs/en/plugins/dependencies)
- [Context7 plans](https://context7.com/plans)
- [Adspirer pricing](https://www.adspirer.com/pricing)
