Corey Corey
Start with Corey
GuidesHow it worksWhat is insidePricingDocsStart with Corey
Corey × Codex

The same operator, a different chair.

Corey was built by Claude, on Claude. It also runs perfectly well in Codex, because Corey is a remote MCP server rather than a plugin. One config block, one browser sign-in, and it is the same operator with the same memory.

Start with Corey →Add Corey to Codex

28 days on us, no card · Connects over MCP · No token to paste

Nothing to rebuild

Changing client changes where you type. That is all.

Your memory comes with you

Corey holds your customers, numbers, decisions and history. None of that lives in the client, so opening Codex does not mean starting again or re-briefing anyone.

The same tools and the same guardrails

Every skill Corey has in Claude it has in Codex, and the four approval gates hold identically: publishing, money, refunds, and anything irreversible still wait for you.

Run it alongside, not instead

Adding Corey to Codex does not remove it from anywhere else. Plenty of people keep it in the CLI at their desk and in Codex when they are already there.

Two minutes

Add Corey to Codex.

Codex adds stdio servers with codex mcp add, but a remote server like Corey goes straight in the config file. Then one command signs you in.

  1. 01

    Add Corey to your Codex config

    Open ~/.codex/config.toml and add an mcp_servers.corey block with the Corey URL. Codex treats a server with a url as a remote streamable-HTTP server, and auth defaults to oauth, so there is nothing else to set.

  2. 02

    Sign in with codex mcp login

    Run codex mcp login corey. Codex opens your browser, you sign in to Corey, and Codex stores and refreshes the tokens for you. There is no access token to paste and no key to keep anywhere.

  3. 03

    Open the folder Corey should run

    Start Codex in the folder you want Corey to manage, for example ~/Corey. Corey reads and writes there natively. Setup is non-destructive, so pointing it at an existing folder is safe.

  4. 04

    Run the kickoff prompt

    Paste the kickoff prompt. Corey confirms the connection, runs the Meet Corey interview one question at a time, writes your workspace, builds your Command Center and shows you three things it can do right now.

~/.codex/config.toml
# auth defaults to oauth - nothing else to set
[mcp_servers.corey]
url = "https://mcp.getcorey.ai/mcp"
sign in
codex mcp login corey
→
No secret goes in that file, so it is safe to commit. Full walk-through, including how to check the connection, in the Corey in Codex doc.
Start with Corey →Other MCP clients
Questions

Before you switch chairs.

Exactly the same one. Corey is a hosted remote MCP server, so every client is a window onto the same operator. Your memory, workspace and standing jobs are identical whichever client you sign in from.
Not if you run Corey in Codex. You need a plan on whichever client you use, so for Codex that is your ChatGPT or API access. The Corey subscription is separate and the same either way: 28 days on us, then £9 every 4 weeks.
Because of what the client adds around Corey, not the quality of Corey inside it. Cursor brings scheduled jobs, cloud agents and an iOS app, which is what lets Corey keep working when you are not at the keyboard. In Codex you drive it yourself.
Yes. Add it in both and sign in to each once. There is no seat to move and no state to sync, because nothing about your setup lives in the client.
No. Corey uses the MCP OAuth flow, so you sign in through your browser and Codex handles the tokens. You never copy a secret into a config file, which is also why the config block below is safe to commit.

Hire Corey where you already work.

The first 28 days are on us, no card. Claude, Cursor, Codex or anything else that speaks MCP - it is the same operator.

Start with Corey →All integrations