Codex plugins extend your coding agent with reusable instructions and connections to external services. They can help Codex review pull requests, implement Figma designs, investigate errors, or deploy through your team's tools.
The best Codex plugins depend on your task and the services your team uses. For an application on Vercel, that might mean combining deployment guidance with GitHub review context. Design work, debugging, and model research need different connections.
Copy link to headingWhat are Codex plugins?
Plugins package reusable skills, MCP server connections, or both. Codex and ChatGPT Work use the same universal plugin directory, although individual capabilities can vary by client.
The components have different jobs:
Installing a plugin can configure a connection, but account access still requires the appropriate authentication and permissions. An MCP server can expose multiple tools; it isn't limited to one API operation.
Agent Plugins provides a portable format for skills and MCP configuration. Commands, hooks, and installation flows can remain client-specific, so a shared package doesn't guarantee identical behavior in every agent.
VS Code extensions operate at the editor level. They can add panels, language support, or an agent interface. Codex plugins extend the agent's available capabilities and can be useful whether you work in an editor or a terminal.
Copy link to headingHow do you install and manage Codex plugins?
In the Codex desktop app, open Plugins from the sidebar, choose a plugin, and follow the installation and connection prompts.
Copy link to headingInstall plugins in the Codex CLI
Start Codex from your terminal:
codexThen enter this command inside the Codex session:
/pluginsChoose the plugin, complete any account setup, and start a new chat before using it. Plugin availability can depend on your plan and workspace settings. Some plugins use their own installer or marketplace, as shown below.
Copy link to headingConnect accounts and manage plugins
Follow the service's authentication instructions when connecting an account. Some integrations use OAuth sign-in; direct MCP connections may require an API key or bearer token.
For token-based HTTP connections, Codex can read credentials from environment variables. Keep secret values out of shared plugin files.
Use the plugin browser to manage or uninstall plugins. To inspect the marketplaces configured for the CLI, run:
codex plugin marketplace listFor separately configured MCP servers, use these commands to list connections and inspect management options:
codex mcp listcodex mcp --helpCopy link to headingWhich Codex plugins fit your workflow?
These six options cover distinct development tasks. The list includes Vercel's own plugin alongside tools for repository work, monitoring, design, project coordination, and AI research. Choose the services you already use or need for the task.
Copy link to heading1. Vercel for platform guidance and deployment
The Vercel Plugin gives Codex guidance for building on Vercel, including AI SDK integration and deployment workflows. Its account connection can inspect projects, deployments, and logs after authentication.
For Codex, the plugin installer selects a native Vercel package from OpenAI's curated marketplace. It omits upstream lifecycle hooks that Codex doesn't support.
With Codex installed and Node.js 18 or later available, use your package manager to run the installer. The target flag keeps installation scoped to Codex:
pnpm dlx plugins add vercel/vercel-plugin --target codexStart a new Codex session in your project, then try a task such as:
Review this app's deployment configuration. Identify any missing environment variables and propose the changes needed for a preview deployment.
Connect the Vercel account when the task needs project or deployment data. Review the target team and project before a deployment or configuration change. Plugin installation doesn't replace those account permissions.
Copy link to heading2. GitHub for repository and pull request work
OpenAI's GitHub plugin connects Codex to repository context for understanding code and preparing changes for review. Its available operations depend on the connected account and tools.
Give Codex a specific pull request or repository and a result to produce:
Review authentication changes in this pull request. Trace affected request paths, identify missing tests, and draft comments for review.
The plugin is useful when a local checkout doesn't contain the discussion or repository history needed for the task. Repository access and permission to publish changes still need to be configured through the connection.
GitHub's standalone MCP server has separate configuration. Its lockdown mode filters some public-repository content to reduce prompt-injection risk. This best-effort filter cannot guarantee prevention or restrict credential permissions.
Copy link to heading3. Sentry for instrumentation and production debugging
Sentry's Codex plugin combines setup and debugging skills with the Sentry MCP server. It can help add instrumentation, investigate reported errors, or configure monitoring. Use the Codex distribution for installation; the separate skill source repository produces the installable packages.
Add Sentry's marketplace, then install its plugin from your terminal:
codex plugin marketplace add getsentry/plugin-codexcodex plugin add sentry@sentry-plugin-marketplaceComplete the Sentry connection and start a new chat. The hosted MCP server uses OAuth; self-hosted Sentry has a separate setup through the Sentry CLI.
For a reported production error, try:
Investigate this Sentry issue and trace it to the relevant code. Explain the likely cause, propose a regression test, and show the patch before resolving the issue.
The debugging workflow can request root-cause analysis from Sentry's Seer when your account supports it. Useful diagnosis still depends on the recorded event data, so missing stack traces or release context can limit what Codex can establish.
Copy link to heading4. Linear for issue context and project coordination
The Linear plugin brings issues and comments into Codex so implementation can follow the team's recorded requirements. Use it to understand a feature's current state or prepare updates that connect code changes to tracked work.
For example, give Codex an issue link and ask:
Read this issue and its discussion. Compare the acceptance criteria with the implementation, then draft a status update explaining what remains.
Check which actions your connection exposes before asking it to change project records. Linear also provides a hosted MCP server with a separate setup and an optional read-only endpoint. Its tools extend beyond basic issue management to initiatives and project milestones.
Adding Linear to Codex gives the agent Linear context. Delegating issues to a coding agent inside Linear requires a separate setup.
Copy link to heading5. Figma for implementing and refining designs
The Figma plugin combines design-to-code and Code Connect skills with service access. Use it to apply layouts and component choices in your codebase.
Share a specific frame and explain how it should fit your application:
Implement this frame using the existing components and design tokens in this repository. Identify any missing component states before adding new styles, then check the result at mobile and desktop widths.
The Figma MCP server can also support native canvas changes and converting live interfaces into editable designs. Canvas access depends on the client and the tools exposed by your connection.
Figma access limits depend on your plan and seat. The connection can only access permitted files, and some image-generation or Figma Make actions use AI credits. Check those limits before using it for repeated design iterations.
Copy link to heading6. Hugging Face for model research and Hub workflows
The Hugging Face plugin helps Codex search models and datasets, inspect repository files, and find relevant research. Use it to choose candidates for an AI feature based on their model cards and licenses.
Give the agent criteria it can check in a focused research request:
Compare multilingual embedding models for this feature using model cards, licenses, and language coverage. Propose an evaluation with our representative queries. Get approval before downloads or compute runs.
The connection can do more than discovery. Supported account actions include working with repositories and invoking Spaces; the Hugging Face MCP server also supports remote Jobs and sandboxes.
Private resources require sign-in and appropriate permissions. Remote compute may incur charges, so review the workload and account limits before launching it. Search results and model cards help narrow a shortlist, but your own evaluation determines whether a model fits the application.
Copy link to headingHow should you combine plugins for a project?
For a new feature, use the Google Drive plugin in ChatGPT Work to prepare an implementation brief from your product requirements. Give Codex the reviewed brief and local repo, then ask it to identify the code changes and tests needed.
For a production bug, combine the failing Sentry event with the relevant GitHub pull request discussion. Codex can use that context to inspect local code, propose a fix, and run the repository's tests.
For a Vercel application, use the Vercel Plugin to review deployment requirements and prepare a preview. Check the behavior in that preview before asking Codex to publish a pull request update or change an issue's status.
If a fix depends on a library API, use the Context7 plugin to retrieve documentation and code examples. Specify the library version so Codex can look for the API your project uses.
Ask for a result you can verify, such as a patch with a passing regression test. Specify the repository, environment, and permitted external changes.
If the agent cannot complete a task, check the connection, permissions, and tools before adding plugins. Overlapping integrations can expose the same actions without providing the missing access or information.
Copy link to headingRun Codex through Vercel AI Gateway
You can route Codex's model requests through Vercel AI Gateway to choose supported models and monitor token usage and spend.
The Vercel Plugin also gives Codex guidance on using AI Gateway in your application, including provider routing, fallbacks, and cost tracking. Ask it to review an existing integration or help configure a fallback when a model request fails.
Configure the connection with the Vercel CLI:
pnpm dlx vercel@latest ai-gateway setup --agent codexThe command configures Codex's model provider and API key, backing up the files it changes. Add --dry-run to preview those changes first.
Start Codex and send a request, then confirm it appears in your AI Gateway Overview. Model usage is billed separately from your ChatGPT subscription.
Copy link to headingCan you share your own plugin with a team?
Package a recurring workflow when several projects need the same instructions. For example, a deployment-review skill could check required environment variable names, identify the build command, and verify the preview URL.
Use plugin.json at the package root, with custom skills in skills/ and optional MCP configuration in mcp.json. Follow the Agent Plugins walkthrough to build the package and separate client-specific behavior.
Existing packages using the older Codex manifest directory remain supported, but that directory is no longer the only packaging option.
Copy link to headingShare a curated plugin list
Register packages in a repo or personal marketplace to share a curated plugin list. Each plugin entry points to its package source, so one catalog can contain several plugins.
Use a repo marketplace for a project-specific list or a personal marketplace for plugins you use across projects. Codex accepts Git repositories and local directories as marketplace sources. Keeping a plugin folder on disk alone doesn't complete installation.
Keep credentials out of the package. Share instructions and connection configuration, then let each user authenticate through the appropriate service.
Public distribution goes through OpenAI's plugin submission and review process. Remove lifecycle hooks before submitting to its public plugin directory.
Copy link to headingFrequently asked questions
Copy link to headingHow do Codex plugins differ from skills and MCP tools?
Codex skills provide reusable instructions and resources for a task. Plugins can bundle related skills with an MCP server connection that supplies tools or external data. Use a skill when instructions and existing tools are enough, or package skills and service connections together when the workflow needs both.
Copy link to headingCan I use Codex plugins in Claude Code?
Some plugins support both Codex and Claude Code, but installation steps and supported features can differ. The Vercel Plugin provides installation targets for both clients. Check the instructions for your client before assuming that bundled skills, MCP tools, or lifecycle hooks behave identically.
Copy link to headingAre plugins supported in ChatGPT Work?
Yes. ChatGPT Work supports plugins for reusable skills and app integrations such as Google Drive and GitHub. ChatGPT and Codex share a public plugin directory, although available capabilities can differ between clients. Plugins marked Desktop only require the ChatGPT desktop app.
Copy link to headingAre Codex plugins free to use?
Installing a plugin doesn't remove the costs or limits of the services it connects. Access can depend on your Codex plan and workspace settings, while connected services may charge for seats, AI features, or remote compute. Check the requirements for the task you plan to run.
Copy link to headingHow do I choose the best Codex plugins for my project?
Choose plugins for the tasks and services your project uses. The GitHub plugin supplies repository and pull request context, while Sentry helps investigate production errors. Enable the integrations needed for your current work and check whether Codex selects the right tools before adding more.
Copy link to headingDo Codex plugins require API key authentication?
No. The authentication method depends on the connected service. Some integrations use OAuth sign-in; others require an API key or bearer token. For HTTP MCP connections that use secret tokens, Codex supports reading credentials from environment variables so you can keep them out of shared configuration files.
Copy link to headingWhy can't Codex use a plugin after installation?
Start a new chat, confirm the plugin is enabled, and complete any connection prompts. Check whether the connected account can access the target resource and whether workspace policy allows the plugin.
If a newly added local plugin is missing, verify its marketplace entry and restart the desktop app. For a separately configured MCP server, use codex mcp list to inspect its configuration.
Reference Architecture
Choose a stack and work through the decisions behind it, with curated documentation, guides, templates, and courses to help you build.
- Agent stack: Build and operate agents with models, tools, and durable workflows.
- Web applications: Choose rendering, data, authentication, and caching approaches.
- Backends: Choose a runtime, connect services, and run background work.
Build with prompts
Build an app, extend an agent, or improve your project with ready-to-use prompts for your coding agent.