OpenAI Group PBC today quietly updated its Codex programming assistant with a feature that enables users to create custom plug-ins.

The launch comes about five months after rival Anthropic PBC released a similar feature for Claude Code. Plugins enable users to extend the tools’ default feature sets with specialized capabilities. Such customizations can speed up project-specific tasks that would otherwise be difficult to automate.

A Codex plugin can include multiple components. Users can equip a plugin with so-called skills, which are workflows designed to automate relatively narrow tasks, and integrations with external applications.

A skill comprises natural language instructions and technical assets such as a script. For example, a developer could upload a firewall configuration script along with an explanation of when Codex should run it. A prepackaged script removes the need for Codex to generate the necessary code from scratch every time it’s needed, which reduces the risk of hallucinations and lowers inference costs.

Plugins can also integrate Codex with external services using MCP servers. Developers may upload configuration files to customize how those MCP servers work. For example, an engineer could connect Codex to an MCP-powered development environment and specify what middleware should be pre-installed in the sandbox.

OpenAI offers more than a dozen prepackaged integrations through a new plugin directory. Those connectors enable Codex to perform tasks such as editing files in Google Drive and reviewing changes to a GitHub repository.

According to OpenAI, a future update will introduce the ability to equip plugins with other types of components besides MCP integrations and skills. Anthropic’s Claude Code enables developers to include automation workflows called sub-agents in their plugins. It’s possible OpenAI plans to match that capability. 

A sub-agent is a version of Claude Code optimized for a specific set of tasks. Users can specify the large language model that a sub-agent should use, provide it with instructions and give it access to third-party applications.

OpenAI says that its plugins are designed to ease collaboration for software teams. When multiple engineers are working on the same project, their development tools must be configured the same way to avoid code inconsistencies. Plugins make it relatively simple to sync a Codex configuration across multiple developers’ OpenAI accounts.

In the future, Codex’s plugin features may also have applications beyond software development.

Earlier this year, Anthropic rolled  out a productivity tool called Claude Cowork. It’s a version of Claude that can use customer-created plugins to automate not only coding tasks but also other use cases. Anthropic has released more than a dozen prepackaged plugins focused on fields such as marketing and finance.

OpenAI reportedly plans to merge Codex and ChatGPT into a “superapp” later this year. If the company decides to carry over Codex’s plugin features into the superapp, it may expand their focus beyond programming to other use cases.

Plugins could potentially also enhance the capabilities of the AI research assistant that OpenAI is reportedly preparing to launch. The ability to add plugin-powered customizations could make it easier for scientists to align the tool’s output with project requirements. 

Image: OpenAI