Codex MCP Server Integration - Setup Guide¶

Source: reason-machines/mcp-skills (129+ installs) Category: Agent Infrastructure / MCP Integration Quality Tier: 🟡 Beta

Bridges OpenAI's Codex CLI into MCP-compatible editors (Claude Code, Cursor, Hermes Agent) through the Model Context Protocol. Provides AI-powered code analysis, generation, review, and web search capabilities without leaving your editor. Architecture: Editor → Codex MCP Server → Codex CLI → OpenAI API.

Installation¶

npx skills add reason-machines/mcp-skills --skill codex-mcp-server-integration

Step 1: Install Codex CLI¶

npm install -g @openai/codex
# Verify: codex --version (0.75.0+ required)

Step 2: Authenticate¶

codex login --api-key "$OPENAI_API_KEY"
codex ping  # Verify

Step 3: Install MCP Server¶

# Claude Code
claude mcp add codex-cli -- npx -y codex-mcp-server

# Or manual MCP config
# Add to MCP settings file (see below)

Prerequisites¶

Requirement	Details
Node.js	18+
Codex CLI	v0.75.0+ (`npm install -g @openai/codex`)
OpenAI API Key	For Codex authentication
MCP-Compatible Editor	Claude Code, Cursor, or Hermes Agent

Key Capabilities¶

Code Analysis & Generation¶

Main codex tool: ask questions about code, request refactoring, generate new implementations. Supports file context arrays and session resumption for multi-turn conversations.

Code Review (`review` tool)¶

Review uncommitted changes, branch diffs, or specific commit ranges. Supports uncommitted, base/head branch comparison, and commit range modes.

Web Search (`web_search` tool)¶

OpenAI-powered web search integrated into the code review flow. Search documentation, find examples, and verify implementation patterns.

Model Selection¶

Override model per-request: o3, gpt-4, gpt-4-turbo. Reasoning effort control for o-series models: low, medium, high.

Sandbox Modes¶

Control execution boundaries: read-only (safe for review), workspace-write (needed for generation), or full autonomous (fullAuto).

Quick Start¶

# Install and configure
npm install -g @openai/codex
codex login --api-key "$OPENAI_API_KEY"
claude mcp add codex-cli -- npx -y codex-mcp-server

# Usage examples (from MCP-compatible editor):
# "Use codex to review my uncommitted changes"
# "Analyze src/auth/login.ts and suggest improvements"
# "Use codex web_search to find React 19 migration docs"

MCP Configuration (Manual)¶

{
  "mcpServers": {
    "codex-cli": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "codex-mcp-server"],
      "env": {
        "CODEX_MCP_CALLBACK_URI": "http://localhost:8080/mcp-callback"
      }
    }
  }
}

Verification¶

# Verify Codex CLI
codex --version
codex ping

# Verify MCP server registration
claude mcp list | grep codex

# Test with a simple prompt
codex "explain this: const x = [1,2,3].map(n => n * 2)"

Notes¶

Codex 0.87+ returns threadId for session resumption - enables multi-turn conversations
Sandbox mode read-only is recommended for code review; use workspace-write only for generation tasks
The review tool is separate from codex - it wraps git diff analysis with AI review
Requires active Codex CLI authentication - tokens expire and need refresh
Web search capability depends on Codex CLI provider configuration (requires OpenAI API)

Integration with CorpusIQ¶

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects