Connect Skillry MCP in Codex (CLI, ChatGPT App and IDE Extension)

The short answer

1Run codex mcp add skillry -- npx --yes skillry-cli@latest mcp once; on Windows wrap the launch in cmd /c.
2Confirm with codex mcp list, or type /mcp in the Codex TUI.
3The ChatGPT desktop app and the IDE extension share the entry: set it up once, use it in all three.

Start here:Add it Check it Authorize It isn’t working

Before you start

Checked against the official documentation on 28 October 2026. Not yet run on a real installation of this client; if a step differs on your machine, use Report an issue at the bottom.

Codex signed in: the CLI, the ChatGPT desktop app with Codex, or the IDE extension. They share one configuration.
Node.js 22 or newer (node --version). Codex starts the server with npx.
A project-level .codex/config.toml also works, but only in projects you have marked as trusted. This guide uses the global file.

Add Skillry to Codex

Fastest when you use the Codex CLI

1
Run this in any terminal:
macOS
```
codex mcp add skillry -- npx --yes skillry-cli@latest mcp
```
You should see: No error, and codex mcp list now includes skillry.
2
On Windows, run this instead so Windows can start npx:
Windows (PowerShell or Command Prompt)
```
codex mcp add skillry -- cmd /c npx --yes skillry-cli@latest mcp
```
3
Start Codex again, or restart the app or extension if one is open.
You should see: skillry in the /mcp list inside Codex.

No terminal needed

1
In the ChatGPT desktop app open Settings, then MCP servers. In the IDE extension open the gear menu, then MCP servers.
2
Select Add server. Enter the name skillry, choose STDIO, and enter the command npx --yes skillry-cli@latest mcp (on Windows cmd /c npx --yes skillry-cli@latest mcp).
3
Save, then select Restart in the app or Restart extension in the IDE.
You should see: skillry listed as enabled in the MCP server list.

When you keep your Codex settings in the file

1
Open ~/.codex/config.toml in an editor; create it if it does not exist.
2
Append this table. If a [mcp_servers.skillry] table already exists, replace it.
macOS
```
[mcp_servers.skillry]
command = "npx"
args = ["--yes","skillry-cli@latest","mcp"]
```

On Windows use this table instead.

Windows

[mcp_servers.skillry]
command = "cmd"
args = ["/c","npx","--yes","skillry-cli@latest","mcp"]

4
If the first start times out, add startup_timeout_sec = 60 under the table. The default is 10 seconds, and the first npx run downloads the package.
5
Save, then start Codex again.
You should see: skillry in codex mcp list and in /mcp.

Check that Codex is connected

1
In a terminal, list the configured servers:
Terminal
```
codex mcp list
```
You should see: skillry with the npx command you entered.
2
In the Codex TUI, type /mcp.
You should see: skillry among the active servers, with its tools.

Try your first search

Searching needs no account, so this is the quickest way to prove the connection works.

1
Start a new conversation in Codex and paste this request.
Prompt your agent
```
Use the Skillry MCP connected to https://skillry.dev to find three product presentation Skills. Show their preview links, required inputs, dependencies and external costs. Do not install or run anything.
```
You should see: Three Skill candidates, each with a preview link, the inputs it needs, its dependencies and any outside costs. Nothing is installed.
2
If the agent answers that it has no Skillry tool, or the request hangs, go to It isn’t working.

Authorize when you install

Installing checks your Skillry account. You do not type a password or paste a key into the agent: the first time, it opens a page in your browser and you approve it there.

1
Ask Codex to install the Skill you chose, for example Install it for this project.
You should see: The agent reports that it needs to connect your account and opens your browser. In a terminal client it may print the link instead; open it yourself.
2
Check that the device code on the page matches the one the agent shows, sign in to Skillry if asked, then select Approve.
You should see: Authorization approved, with a note to return to your agent. The agent notices within a few seconds and finishes the install.

The approval lasts 7 days and is stored in your operating system’s keyring (Keychain on macOS, Credential Manager on Windows). When it expires the agent simply asks you to approve again. To sign out early, ask the agent to disconnect your Skillry account.

If Codex can’t see Skillry

What you see	Likely cause	What to do
`skillry` is missing from `codex mcp list`	The table went into a project file in an untrusted project, or the table name has a typo	Put the table in `~/.codex/config.toml` and check it reads exactly `[mcp_servers.skillry]`.
The server never becomes ready, or Codex reports a startup timeout	The first `npx` download took longer than `startup_timeout_sec` (10 seconds by default)	Add `startup_timeout_sec = 60` to the table, or run the launch command once by hand so the package is cached.
Works in the CLI but not in the ChatGPT app	The app was open while the file changed	Quit and reopen the ChatGPT desktop app, or select Restart extension in the IDE.
The server is listed but shows as failed, or never starts	`npx` is not installed, Node.js is older than 22, or the first start timed out while downloading the package	Run `node --version` and `npx --version` in a terminal. Then run `npx --yes skillry-cli@latest mcp` once by hand, wait for it to start, press Ctrl+C, and try the client again: the package is now cached.
On Windows the server exits as soon as it starts	The entry starts `npx` directly, and Windows needs a shell to run it	Use the Windows entry from this page: the command is `cmd` and the arguments start with `/c`, `npx`.
The browser page says the request is invalid or expired	The device code was approved too late, or belongs to an earlier attempt	Ask the agent to connect again and approve the new code within a few minutes.
Installing stops with a Premium or purchase message	The Skill is Premium and your account has neither a plan nor a purchase for it	Follow the link the agent shows to buy the Skill or subscribe, then ask again. Nothing is charged until you do.

Remove or disconnect

1
Delete the [mcp_servers.skillry] table from ~/.codex/config.toml, or use the remove command that codex mcp --help lists for your version. To switch it off without deleting it, add enabled = false to the table.
2
Restart Codex, the app or the extension.
You should see: skillry gone from codex mcp list.

Removing the entry does not sign you out of Skillry. To revoke the stored approval as well, ask the agent to disconnect your Skillry account first, or remove the entry and let the approval expire after 7 days.

Reference

Item	macOS	Windows
Global config file	`~/.codex/config.toml`	`%USERPROFILE%\.codex\config.toml`
Project config file	`.codex/config.toml` (trusted projects only)	same
Launch command	`npx --yes skillry-cli@latest mcp`	`cmd /c npx --yes skillry-cli@latest mcp`
Check	`codex mcp list` or `/mcp`	same

The table Codex stores (macOS)

[mcp_servers.skillry]
command = "npx"
args = ["--yes","skillry-cli@latest","mcp"]

Timeouts, tool allow-lists and approval modes are in the official Codex MCP documentation.

Common questions

Do I set it up separately for the ChatGPT desktop app?

No. The app, the CLI and the IDE extension read the same config.toml. Restart the app after changing it.

Can I use it in one project only?

Yes. In a trusted project, put the table in .codex/config.toml inside that project instead of the global file.

Why is the first start slow?

The first npx run downloads skillry-cli. Raise startup_timeout_sec or run the launch command once by hand; after that it starts at once.

Sources

Did this guide help?

PreviousClaude Code NextCursor