OpenAI Codex CLI Plugin — Cromus Workflow Intelligence | Cromus

Cromus ships an HTTP MCP plugin for OpenAI Codex CLI. The plugin exposes fifteen deterministic Workflow Intelligence tools, including SKILL.md and graph scoring, bounded graph path simulation, validation, drift checks, policy checks, and approved-skill retrieval, as standard MCP tools that Codex can call from the terminal.

Install is a single config block in ~/.codex/config.toml (or ~/.codex/mcp.json on older Codex builds) pointing at https://mcp.cromus.ai/mcp with a Bearer token generated at https://cromus.ai/app/settings/api-keys. There is no local process to run. The plugin shares the same /mcp endpoint and key system as the Claude Desktop, Claude Code, and Claude Cowork connectors, so one key works everywhere.

Codex excels at writing and refactoring code, but it cannot tell on its own what a skill costs or whether it violates an ETHOS.md policy. The Cromus plugin closes that gap by giving Codex pre-execution Workflow Intelligence as MCP tools. Every tool result is deterministic, reproducible, and carries attribution metadata. The plugin is independent and not affiliated with OpenAI.