---
title: "Codex app | OwlMeans"
description: "Connect OwlMeans to a local project in the Codex app. Make sure the app can launch npx from its environment."
url: "https://owlmeans.com/docs/mcp/clients/codex-app/"
lang: "en"
---

# Codex app

![](https://owlmeans.com/_astro/mcp-clients-codex-app.JDfqbPbt_Z1KtC0z.webp)

Connect OwlMeans to a local project in the Codex app. Make sure the app can launch `npx` from its environment.

1.  Open the project in the app.

2.  Open the app’s MCP settings and add `owlmeans` as a stdio server, or edit its supported Codex configuration.

    ![OwlMeans MCP settings: 1, inference preference; 2, client setup tabs.](https://owlmeans.com/_astro/mcp-settings.C46Z_RMB_zICmB.webp)

    OwlMeans MCP settings: 1, inference preference; 2, client setup tabs.

3.  Use command `npx` and the argument array below.

4.  Save, refresh the connection and begin a new project conversation.

```
[mcp_servers.owlmeans]
command = "npx"
args = ["-y", "@owlmeans/viable-mcp@^0.2.0", "--harness", "codex"]
startup_timeout_sec = 90
tool_timeout_sec = 300
```

Codex’s MCP configuration uses `mcp_servers` tables. A desktop app launched outside a terminal may have a different PATH; use the resolved executable path when `npx` is not found. See [official MCP configuration](https://learn.chatgpt.com/docs/extend/mcp) and the client’s own current settings.

Use a local app session for local files and delegated execution. The [ChatGPT guide](https://owlmeans.com/docs/mcp/clients/chatgpt-web-and-desktop) covers a hosted chat connection.

## Check the connection

1.  Complete OwlMeans browser sign-in when requested.
2.  Ask the agent to call `describe_platform` and `session_status`.
3.  Verify the intended project, workspace and inference mode.
4.  Ask it to inspect or [install the project harness](https://owlmeans.com/docs/mcp/install-and-use-your-project-harness).
5.  Start a small [generation or story workflow](https://owlmeans.com/docs/mcp/workflows/generate-an-app-and-develop-its-user-stories).

The examples use a **local target**. Add `--target`, `cloud` as separate arguments for a hosted project. Add `--llm`, `local` for delegated inference only when the connected coding agent will handle tasks. [Mode selection](https://owlmeans.com/docs/mcp/choose-project-location-and-inference-mode) explains the effects and costs.

Keep access tokens out of committed configuration. Restart the client after changing the server settings. If a tool is absent, inspect the connected server’s tool list and [troubleshoot](https://owlmeans.com/docs/mcp/troubleshoot-your-mcp-connection).

## Names and marks

OpenAI, ChatGPT and related OpenAI marks belong to OpenAI; Codex is an OpenAI product. [Usage policy](https://openai.com/brand/).

Claude and Claude Code are Anthropic products; their names and marks remain with their respective owner. [Usage policy](https://www.anthropic.com/legal/commercial-terms).

Node.js, npx names and any associated marks remain with their respective owners or project maintainers. References identify products, standards or source examples. OwlMeans documentation does not imply sponsorship or endorsement by those owners.
