How to Connect MCP Servers to OpenAI Codex CLI

Codex CLI Supports MCP

OpenAI's Codex CLI is an autonomous coding agent that runs in your terminal. It supports MCP servers natively, letting it call external tools during task execution. When Codex encounters a problem that requires external data (API lookups, database queries, service status), it can invoke MCP tools without leaving the terminal session.

Prerequisites

• Codex CLI installed (npm install -g @openai/codex)
• An MCP server URL (SSE endpoint)
• OpenAI API key configured for Codex

Step 1: Configure MCP Servers

Codex reads MCP configuration from ~/.codex/config.json. Create or edit this file:

{
  "mcpServers": {
    "analytics": {
      "type": "sse",
      "url": "https://datafaucet.dev/api/mcp/YOUR_SERVER_ID/sse",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Replace YOUR_SERVER_ID and YOUR_API_KEY with values from your DataFaucet dashboard.

Step 2: Verify Connection

Run Codex with a simple tool query:

codex "What tools do you have access to?"

Codex should list available MCP tools from your configured servers. If tools don't appear, verify the URL is correct and the API key is valid.

Step 3: Use Tools in Tasks

Once connected, Codex calls MCP tools when relevant to the task:

codex "Check yesterday's error rate and fix any issues in the logging module"
codex "Pull the latest customer feedback and update the FAQ page"
codex "What's the deployment status? If green, push the current branch"

Codex autonomously decides when to invoke tools based on the task context. No explicit tool calls needed in your prompt.

Multiple Servers

Add as many servers as needed:

{
  "mcpServers": {
    "analytics": {
      "type": "sse",
      "url": "https://datafaucet.dev/api/mcp/ANALYTICS_ID/sse",
      "headers": { "Authorization": "Bearer KEY_1" }
    },
    "crm": {
      "type": "sse",
      "url": "https://datafaucet.dev/api/mcp/CRM_ID/sse",
      "headers": { "Authorization": "Bearer KEY_2" }
    },
    "monitoring": {
      "type": "sse",
      "url": "https://datafaucet.dev/api/mcp/MONITORING_ID/sse",
      "headers": { "Authorization": "Bearer KEY_3" }
    }
  }
}

Codex picks the right server based on tool names and descriptions. Naming servers descriptively helps Codex select the correct tool.

Stdio vs SSE

Codex supports both transport types:

• SSE (recommended for DataFaucet): remote, hosted, always available
• stdio: runs locally as a subprocess, starts/stops with Codex

DataFaucet generates SSE servers by default. No local process management required.

Security Notes

Codex operates autonomously. When connected to MCP servers that access production data:

• Use read-only API keys when possible
• Scope tools to specific datasets (DataFaucet lets you select which endpoints to expose)
• Review Codex's tool call log after execution (codex --verbose)

DataFaucet + Codex Workflow

1. Browse any web app with DataFaucet (60 seconds)
2. Deploy the generated MCP server
3. Add the SSE URL to ~/.codex/config.json
4. Give Codex tasks that reference that app's data

Codex handles the rest. It reads tool schemas, decides when to call them, and incorporates results into its task execution. Total setup: under 2 minutes per service.