OpenAI Codex
OpenAI Codex CLI is OpenAI’s command-line coding assistant. This guide shows how to route Codex’s MCP connections through Maybe Don’t.
Prerequisites
- Maybe Don’t running (see MCP overview)
- OpenAI Codex CLI installed
- A GitHub Personal Access Token
Configure Codex
Codex uses TOML configuration at ~/.codex/config.toml. Add the MCP server configuration:
[mcp_servers.maybe-dont]
url = "http://localhost:8080/mcp"
env_http_headers = { "X-GitHub-Token" = "GITHUB_TOKEN" }The env_http_headers field maps header names to environment variable names. Codex reads the value of GITHUB_TOKEN from your environment and sends it as the X-GitHub-Token header.
Set Environment Variables
Before starting Codex, set your GitHub token:
export GITHUB_TOKEN="ghp_your_token_here"Start Codex
Start Codex normally:
codexVerify the Connection
In your Codex session, try asking Codex to list its available MCP tools. You should see GitHub tools prefixed with github__.
Static Headers
If you prefer to hardcode the token (not recommended for secrets):
[mcp_servers.maybe-dont]
url = "http://localhost:8080/mcp"
http_headers = { "X-GitHub-Token" = "ghp_your_token_here" }env_http_headers over http_headers for tokens and secrets. Static headers are stored in plain text.What’s Happening
When Codex calls a tool:
- Codex sends the request to
localhost:8080/mcpwith your GitHub token - The gateway validates the request using configured policies
- If allowed, the gateway forwards to GitHub
- All activity is logged to the audit log
Troubleshooting
Connection issues
Verify the gateway container is running:
docker ps | grep maybe-dontCheck the gateway logs for tool discovery messages or errors:
docker logs maybe-dontAuthentication errors
Verify your GitHub token works:
curl -H "Authorization: Bearer $GITHUB_TOKEN" \
https://api.github.com/user