Partial MCP Support

Tools for OpenAI Codex

Extend Codex with the same powerful MCP tools used by Cline and Claude Code. Connect through the community adapter layer and unlock 80+ tools for your coding agent.

MCP with Codex

Codex supports MCP tools through a community-built adapter that bridges the OpenAI function-calling API with MCP servers.

How It Works

The MCP-to-Codex adapter runs as a lightweight bridge process. It starts your MCP servers locally, discovers their available tools, and translates them into OpenAI-compatible function definitions that Codex can call natively.

From Codex's perspective, MCP tools look like standard functions. From the tool's perspective, it is being called via the normal MCP protocol. The adapter handles all translation transparently.

Note: Since Codex runs tasks in sandboxed cloud environments, the adapter connects MCP servers as remote endpoints. Local-only tools (like filesystem access) are scoped to the Codex sandbox, not your host machine.
1

Install the adapter

Install the MCP-Codex bridge package globally via npm.

2

Configure your MCP servers

Add MCP server definitions to the adapter config file.

3

Start the adapter

Run the bridge process, which starts MCP servers and exposes tools to Codex.

4

Use tools in Codex

Codex automatically discovers and uses the available MCP tools.

Terminal
# Install the MCP-Codex adapter
npm install -g @clinetools/mcp-codex-adapter
mcp-codex.config.json
{
  "servers": {
    "secret-scanner": {
      "command": "npx",
      "args": ["@clinetools/secret-scanner"]
    },
    "git": {
      "command": "npx",
      "args": [
        "@modelcontextprotocol/server-git"
      ]
    },
    "complexity": {
      "command": "npx",
      "args": [
        "@clinetools/complexity-analyzer"
      ]
    }
  },
  "codex": {
    "autoDiscover": true,
    "timeout": 30000
  }
}
Terminal
# Start the adapter bridge
mcp-codex-adapter start --config ./mcp-codex.config.json

# Output:
# MCP-Codex Adapter v1.2.0
# Loading 3 MCP servers...
#   secret-scanner: 3 tools ready
#   git: 8 tools ready
#   complexity: 2 tools ready
# Bridge active - 13 tools available to Codex

Recommended Tools for Codex

These 10 tools are tested and verified to work seamlessly with the Codex adapter. Start here for the best experience.

1

Secret Scanner

Scan code for leaked API keys, tokens, and credentials before they reach your repository.

Security Verified
2

Git Server

Full Git operations including status, diff, commit, branch, and log without shell access.

Infrastructure Official
3

Complexity Analyzer

Measure cyclomatic complexity and identify functions that need refactoring.

Code Quality Verified
4

Filesystem Server

Sandboxed file read, write, and search operations within the Codex workspace.

Filesystem Official
5

Dead Code Detector

Find unused exports, functions, and variables across JavaScript and TypeScript codebases.

Code Quality Verified
6

Prompt Detector

Analyze text inputs for prompt injection patterns and adversarial content.

Security Verified
7

PostgreSQL Client

Run queries, inspect schemas, and manage database state directly from your agent.

Database Verified
8

Docker Manager

Build images, run containers, and inspect logs without direct Docker CLI access.

Infrastructure Verified
9

Type Coverage

Measure and report TypeScript type coverage across your codebase with detailed reports.

Code Quality Verified
10

Web Fetcher

Fetch and parse web pages, APIs, and documentation from within the agent context.

Data Official

Coming Soon

The roadmap toward full, native MCP integration with OpenAI Codex. Here is where things stand and what is next.

Phase 1
Now

Community Adapter

The current approach uses a community-built adapter to bridge MCP servers with Codex's function-calling interface. Works well for most tools with minor latency overhead.

MCP-to-function-call translation
Multi-server support
80+ tools verified working
Phase 2
Q2 2026

First-Party Adapter

An officially maintained adapter with improved performance, error handling, and streaming support. Reduced setup friction with one-command installation.

Streaming tool responses
One-command install and setup
Better error propagation
Phase 3
TBD

Native MCP Support

Full, built-in MCP support directly in the Codex runtime. No adapter needed. Same seamless experience as Cline and Claude Code, with tools automatically discovered and available.

Zero-config MCP server loading
Native resource and prompt support
Full parity with other MCP-native agents

Start using MCP tools with Codex today

Browse the full catalog to find tools for your workflow, or check the adapter setup guide to get started in minutes.