> ## Documentation Index
> Fetch the complete documentation index at: https://flow.engineer/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Codex plugin

> Package your product's tools and skills for Codex.

Use Flow's Agent Plugins output for Codex. The package contains workflow skills and a connection
to your tools; it does not start the MCP server.

## Prepare the package

```sh Terminal
npx @flow-engineer/flow plugin --server https://agent.example.com/mcp --out plugins
npx @flow-engineer/flow check --plugin plugins
```

Use the generated `plugins/agent-plugins/` folder:

```text Package
plugin.json
mcp.json
skills/
  <name>/SKILL.md
```

## Install and test

**1. Make the server reachable**

Keep the MCP server running. Use loopback for a local client or public HTTPS for a remote
environment. Configure user sign-in before exposing account-protected tools.

**2. Add the plugin to a supported source**

Follow [OpenAI's plugin packaging guide](https://developers.openai.com/plugins/build/plugins)
for local or team distribution. OpenAI-specific settings belong under `extensions.com.openai`
in the portable manifest.

**3. Check a workflow**

Install the plugin, open a new session and ask Codex to perform a task covered by a skill.

**Check it worked:** the skill is available and its tool calls reach `/mcp/plugin`.

Regenerate after changing skills, then refresh the installed package. Flow does not add a
marketplace entry or publish to a directory. For direct tool use without a plugin, connect the
[MCP server](https://flow.engineer/docs/framework/mcp.md) or use the [CLI](https://flow.engineer/docs/cli.md).