# Codex CLI as an MCP server (retired): an MCP server we run in production

Canonical: https://scalably.io/mcp/codex-cli-mcp
Source: https://github.com/openai/codex
Registry: not published to the MCP registry
Provenance: third-party server, pinned at codex mcp-server, verified 2026-09-09. Verification note: codex mcp-server answered initialize up to codex-cli 0.153.x; removed in 0.154.0
Integrity: no independent hash is published for this third-party listing; the raw endpoint at https://scalably.io/mcp/codex-cli-mcp.md mirrors this representation byte for byte
This is the machine-readable representation of the page at the canonical URL. Same facts, denser format.

## Retired 2026-10-02
OpenAI removed the `codex mcp-server` command and the standalone `codex-mcp-server` binary from the Codex CLI on 2026-09-05 (openai/codex PR #42993, first stable release without it: 0.154.0). OpenAI's docs now read: "Integrations that launch either command must migrate before upgrading Codex." The suggested replacement, the Codex app server, isn't an MCP server and is documented as experimental, so we no longer run Codex as an MCP tool. The tools below are kept for reference as they existed up to 0.153.x: `codex` (start a session with a prompt, model, sandbox and approval policy) and `codex-reply` (continue a session by thread id).

Current servers: https://scalably.io/mcp/

## Direct answer
Was a second implementer and reviewer reachable as a tool from Claude Code, until OpenAI removed codex mcp-server.

## Install
- Claude Code: `claude mcp add codex -- codex mcp-server`
- Codex: not available for this server.
- Claude Desktop: not available for this server.

## Environment
No environment variables.

## Tools
| Tool | Description |
|---|---|
| `codex` | Up to 0.153.x: run a Codex session from a prompt, with optional model, cwd, sandbox (read-only, workspace-write, danger-full-access) and approval policy. |
| `codex-reply` | Up to 0.153.x: continue a Codex session by thread id with a new prompt. |

## Setup
No environment variables required. Follow the vendor's own install instructions.

## Limits
Not documented by OpenAI. See the vendor's own documentation: https://github.com/openai/codex

## Reliability
- CI: not applicable, this is a third-party project we do not control.
- Pinned at codex mcp-server
- Verified 2026-09-09
- We run this server in production, pinned at the exact version, with the install command derived mechanically from that pin so it can be re-verified against the published package. codex mcp-server answered initialize up to codex-cli 0.153.x; removed in 0.154.0

## Related
- Read more: [Codex skills: build reusable workflows that trigger](https://scalably.io/blog/codex-skills), [Claude Code MCP: how to add an MCP server](https://scalably.io/blog/claude-code-mcp)