Codex CLI

How to use open-mcp-guardrails with OpenAI Codex CLI.

Project-Level Setup

Create .codex/config.toml in your project root:

[mcp_servers.filesystem]
command = "bunx"
args = [
  "open-mcp-guardrails",
  "--",
  "bunx", "@modelcontextprotocol/server-filesystem", "/tmp"
]

Codex reads this file automatically when you start a session in the project directory.

Using the CLI

You can also add MCP servers with the codex mcp add command:

codex mcp add filesystem \
  -- bunx open-mcp-guardrails \
  -- bunx @modelcontextprotocol/server-filesystem /tmp

Scopes

Codex supports two configuration scopes:

Scope	Location	Shared with team
Project	.codex/config.toml (project root)	Yes (commit to git)
Global	~/.codex/config.toml	No

Guarding Multiple Servers

[mcp_servers.filesystem]
command = "bunx"
args = [
  "open-mcp-guardrails",
  "--",
  "bunx", "@modelcontextprotocol/server-filesystem", "/tmp"
]

[mcp_servers.github]
command = "bunx"
args = [
  "open-mcp-guardrails",
  "--",
  "bunx", "@modelcontextprotocol/server-github"
]

[mcp_servers.github.env]
GITHUB_TOKEN = "..."

Environment Variables

Environment variables can be set per-server using the [mcp_servers.<name>.env] table:

[mcp_servers.my-server]
command = "bunx"
args = [
  "open-mcp-guardrails",
  "--",
  "bunx", "my-mcp-server"
]

[mcp_servers.my-server.env]
API_KEY = "..."

Next Steps

• Configuration — Detailed config options
• Rules — Available rule types