Connect Codex (OpenAI CLI)

Prerequisites

• OpenAI Codex CLI installed (npm install -g @openai/codex)
• A Wicket member key — see the Quickstart if you don’t have one yet
• mcp-remote installed (npm install -g mcp-remote) — bridges HTTP MCP servers to Codex’s stdio transport

Configuration

Codex uses stdio-based MCP servers. Use mcp-remote as an adapter for Wicket’s HTTP endpoint.

Edit ~/.codex/config.yaml and add under mcpServers:

mcpServers:
  wicket:
    command: mcp-remote
    args:
      - https://mcp.wicket.sh/mcp
      - --header
      - "Authorization: Bearer <your-member-key>"

Or pass inline when running Codex:

CODEX_MCP_SERVERS='{"wicket":{"command":"mcp-remote","args":["https://mcp.wicket.sh/mcp","--header","Authorization: Bearer <your-member-key>"]}}' codex

Tip

Store your member key in an environment variable to avoid inlining it:

mcpServers:
  wicket:
    command: mcp-remote
    args:
      - https://mcp.wicket.sh/mcp
      - --header
      - "Authorization: Bearer ${WICKET_KEY}"

Verify

Run codex and ask it to use a tool from your connected connector:

> List my GitHub repositories.

Confirm the call appears in the Wicket Audit tab.

Troubleshooting

Symptom	Likely cause
command not found: mcp-remote	Run npm install -g mcp-remote
401 Unauthorized	Wrong or expired member key — regenerate in Wicket
403 Forbidden	Agent disabled or member not approved
Tool call blocked	Policy deny — check the Audit tab