# Connect OpenAI Codex CLI to Zorlya

> Add Zorlya to Codex CLI as a Streamable HTTP MCP server in config.toml, sign in with codex mcp login, and query your matches.

Source: https://zorlya.com/guides/ai-clients/codex-cli

Updated 2026-10-25.

To connect Codex CLI to Zorlya, add the server address `https://mcp.zorlya.com` to `config.toml` and run `codex mcp login zorlya`. Codex CLI is not on Zorlya's list of AI apps, so the connection may be refused. We have not tried it.

**Not on Zorlya's list yet.** Zorlya accepts only the AI apps on its list: Claude, ChatGPT, Claude Code and Cursor ([how it works](/guides/ai-clients/what-you-approve-when-you-connect)). Codex CLI is not on it, so Zorlya refuses its registration unless Codex CLI sends you back to an address that a listed app uses. Its default callback address, `http://127.0.0.1/callback`, is the one Claude Code uses, so the Connect page would name Claude Code. We have not tried it. The steps below are what Codex CLI's own documentation describes for a remote server.

## What you need

- An approved Zorlya account. A job seeker account needs the Pro plan, because AI client access is part of Pro.
- Codex CLI installed. The documentation page does not state a minimum version.
- A browser on the same machine, because the sign-in redirects to a local address.

## Redirect address

Codex's documentation gives the default callback as `http://127.0.0.1/callback`. When the provider does not advertise issuer support, Codex appends a server-specific ID, for example `http://127.0.0.1/callback/XuuuHAzzHOni`. Zorlya's list allows `http://127.0.0.1/callback` on any port for Claude Code, so the first address matches that entry, and the Connect page would then show the name Claude Code. The second, with the server ID after `/callback`, has a longer path and does not match. The page does not say how the port is chosen. To fix a port, it documents `mcp_oauth_callback_port` globally or `mcp_servers.<server-name>.oauth.callback_port` for one server.

## Steps

1. Open `~/.codex/config.toml`. A project can have its own `.codex/config.toml`, but only for trusted projects.
2. Add a table for Zorlya. `url` is the only required key, and `auth` defaults to `oauth`:

   ```toml
   [mcp_servers.zorlya]
   url = "https://mcp.zorlya.com"
   ```

3. Run:

   ```bash
   codex mcp login zorlya
   ```

4. On Zorlya's Connect page, choose the account and the permission level, then sign in. Read only is the default. Choose Read and change only if you want Codex to change your saved jobs and applications.
5. Start `codex`, then type `/mcp` to list active servers.

If the login step reports that registration failed, the documentation lets you choose a method for one login with `--oauth-client-registration` set to `cimd` or `dcr` (the default is `auto`). Zorlya supports dynamic registration, so `dcr` is the one to try.

## What you should see

`/mcp` in the Codex TUI lists `zorlya` among your active servers with its tools. Access tokens last 1 hour and refresh tokens 30 days. The documentation page does not say where Codex stores the tokens, so we cannot tell you.

## Try it

- "Use Zorlya's top_matches to list my 5 best matches and explain the highest one."
- "Search Zorlya for hybrid roles in data engineering and show salary where stated."

## If it does not connect

1. **Not logged in.** Run `codex mcp login zorlya` again. The `bearer_token_env_var` and header keys on the page are for static tokens, and Zorlya has none, so leave them out.
2. **Registration refused.** Zorlya accepts only the return addresses of the AI apps on its list. A callback address other than `http://127.0.0.1/callback` is refused.
3. **No browser on the machine.** The callback is on `127.0.0.1`. On a remote host, Codex's page describes `mcp_oauth_callback_url` for a remote Devbox ingress URL, which Zorlya refuses, because no listed app uses it.

## What we checked

The steps come from Codex's MCP documentation, read on 2026-10-17. We did not try the connection to Zorlya in Codex. The page documents `codex mcp add` with `--url` only in a pre-registered client example, so we show the `config.toml` table instead.

More on the server is on [the developer page](/developers).

## Sources

- https://developers.openai.com/codex/mcp (redirects to https://learn.chatgpt.com/docs/extend/mcp?surface=cli)