OpenAI Codex Plugins Bring Slack, Figma, and Notion to 1.6 Million Developers

OpenAI shipped a plugin system for Codex on March 25, 2026, and within 48 hours pushed a follow-up CLI release (0.117.0) that makes plugins a “first-class workflow.” The move transforms Codex from a coding agent into something closer to a full developer platform — one where you can pull in Slack threads, Figma designs, and Notion docs without leaving your coding environment.

With 1.6 million weekly active users and a Windows launch still fresh, the timing is deliberate. OpenAI isn’t just competing on model quality anymore. It’s competing on ecosystem.

What’s Actually in the Plugin Directory

The Codex Plugin Directory is live now with curated integrations from recognizable names:

Slack — Kick off coding tasks directly from Slack channels and threads. Mention @Codex with a prompt, and it spins up a cloud task and replies with the results.
Figma — The Figma MCP Server connects Codex to Figma Design, Make, and FigJam files. Codex can pull design context and use it as part of the build process.
NotionHQ — Pulls in documentation and project context from Notion workspaces.
Gmail — Email integration for workflows that involve communication alongside code.
Cloudflare — Deploy and manage Cloudflare services directly through Codex.

These aren’t just API wrappers. Each plugin bundles three component types: skills (prompt templates that describe workflows), apps (integration connectors), and MCP server configurations (remote tools and shared context). A single plugin can combine all three into one installable package.

The directory works across the Codex desktop app, the CLI, and IDE extensions for VS Code, Cursor, and JetBrains. Self-publishing is listed as “coming soon” — right now, only curated plugins appear in the official directory.

Under the Hood: How Codex Plugins Work

The plugin architecture borrows from package management patterns that developers already know. Each plugin lives in a structured directory:

my-plugin/
├── .codex-plugin/
│   └── plugin.json    (manifest — required)
├── skills/
│   └── my-skill/
│       └── SKILL.md
├── .app.json          (optional integrations)
├── .mcp.json          (optional MCP config)
└── assets/

The plugin.json manifest defines metadata (name, version, description, author) plus pointers to the plugin’s components. Plugin names follow kebab-case, and the manifest supports marketplace-specific fields like brandColor, composerIcon, and defaultPrompt.

There are two distribution models beyond the official directory:

Repository-level marketplace — Drop a marketplace.json file in your repo’s .agents/plugins/ directory. Every developer who clones the repo gets access to those plugins. This is the team sharing path.

Personal marketplace — Store plugins in ~/.codex/plugins/ with a personal marketplace.json. These are private to your machine.

Installation policies are configurable per plugin: AVAILABLE (opt-in), INSTALLED_BY_DEFAULT (auto-installed), or NOT_AVAILABLE (blocked). Authentication can happen at install time or on first use. The March 26 CLI update added startup syncing for product-scoped plugins, meaning Codex now pulls in plugin updates automatically when you launch it.

OpenAI also ships a @plugin-creator skill that helps you scaffold new plugins from within Codex itself — a meta touch that makes the ecosystem self-bootstrapping.

The Real Competition: Cursor and GitHub Copilot Are Building Ecosystems Too

OpenAI isn’t the first to recognize that plugins matter. The AI coding tool landscape has shifted from “who has the best model” to “who has the best platform,” and all three major players are making moves.

Cursor launched its own Plugin Marketplace with partners including Amplitude, AWS, Figma, Linear, Stripe, Cloudflare, Vercel, Databricks, Snowflake, and Hex. Cursor’s plugins bundle MCP servers, skills, subagents, rules, and hooks — a very similar architecture to Codex plugins. Being a standalone AI-native IDE, Cursor has deeper control over the developer experience. Its tab completions and inline diffs are tighter because everything runs locally.

GitHub Copilot takes the extension route, plugging into VS Code, JetBrains, Neovim, and Visual Studio. Copilot’s cloud agent runs in GitHub Actions VMs and can choose between Claude, Codex, or Copilot as the underlying model. Its competitive advantage is proximity to the GitHub ecosystem — issues, PRs, Actions, and code review are all native.

Where Codex differentiates is scale and pricing. At 1.6 million weekly active users, it has a larger active base than most competitors. And because Codex bundles with ChatGPT subscriptions — Plus at $20/month, Pro at $200/month — there’s no separate billing for the coding agent. Cursor Pro runs $16/month on annual plans, and Copilot Pro is $8.33/month ($100/year). But neither includes a general-purpose AI assistant the way ChatGPT does.

The plugin architecture itself is largely convergent across all three tools. MCP (Model Context Protocol) has become the common standard for connecting AI agents to external tools, and all three platforms support it. The differentiation will come down to which ecosystem attracts more third-party plugin developers — and that’s a numbers game that favors whoever has the most users.

From Coding Agent to “Super App”

The plugin launch doesn’t exist in isolation. It’s part of OpenAI’s broader strategy to merge ChatGPT, Codex, and the Atlas browser into what internal planning reportedly calls a “super app.” Plugins are the connective tissue that makes this possible.

Consider the workflow implications. A product manager writes a spec in Notion. A designer creates mockups in Figma. A developer opens Codex, and the coding agent can now pull context from both sources — the spec and the designs — without anyone copying and pasting between tools. The Slack plugin lets team members trigger Codex tasks from conversations, and results flow back into the same thread.

This is the shift from “AI writes code” to “AI coordinates work.” The plugin system makes Codex useful for planning, research, and coordination tasks that extend well beyond writing functions and fixing bugs.

For enterprise teams, the repository-level marketplace is particularly interesting. A platform engineering team can curate a set of approved plugins — company-specific deployment tools, internal documentation connectors, compliance checkers — and distribute them through the repo. Every developer on the team gets the same setup without manual configuration.

What Developers Are Saying

Community feedback has been cautiously positive. On the OpenAI Developer Community forums, developers highlight Codex’s reliability for complex development work and the lack of hard usage limits on the $20 Plus plan as key advantages over competing tools. The plugin system is seen as a natural extension of what Codex already does well.

The main criticism centers on cloud latency. Because Codex runs tasks in remote sandboxes, individual operations feel slower than Cursor’s local execution. OpenAI compensates with parallelism — you can run multiple Codex tasks simultaneously — but for developers used to instant feedback, the tradeoff is noticeable.

There’s also skepticism about the “walled garden” risk. With ChatGPT, OpenAI launched a plugin marketplace in 2023 that eventually got rolled back. Developers who invested in building ChatGPT plugins were left with deprecated integrations. OpenAI needs to demonstrate that the Codex plugin ecosystem has longer staying power.

Frequently Asked Questions

Is OpenAI Codex Plugins free to use?

Codex plugins are available to all Codex users at no additional cost. Codex itself is included with ChatGPT Plus ($20/month), Pro ($200/month), Business ($30/user/month), and Enterprise plans. For a limited time, Codex is also available on ChatGPT Free and Go tiers with reduced rate limits.

How does OpenAI Codex Plugins compare to Cursor’s Plugin Marketplace?

Both platforms use similar architectures built on MCP (Model Context Protocol) and support integrations with tools like Figma, Slack, and Cloudflare. Cursor’s marketplace launched with more third-party partners (Amplitude, AWS, Stripe, Databricks, Snowflake, Hex) and benefits from deeper IDE integration. Codex’s advantage is its larger user base (1.6 million weekly active users) and bundled pricing with ChatGPT.

Can I build and publish my own Codex plugin?

You can build plugins locally today using the documented plugin structure and distribute them through repository-level or personal marketplaces. Publishing to the official Codex Plugin Directory is not yet self-serve — OpenAI says this capability is “coming soon.”

What tools and integrations are currently available?

The curated directory includes Slack, Figma, NotionHQ, Gmail, and Cloudflare. Since plugins are built on MCP, any existing MCP server can theoretically be wrapped into a plugin. The ecosystem is expected to grow quickly once self-publishing opens up.

Does Codex Plugins work in my IDE?

Yes. Codex plugins work across the Codex desktop app (Mac and Windows), Codex CLI, and IDE extensions for VS Code, Cursor, and JetBrains IDEs. Plugin availability and behavior is consistent across all surfaces.

Top AI Product

Leave a comment Cancel reply