Skip to content

Connect Claude Code, Codex or another tool to Player One

Once a coding tool is connected, you can say to it, "What decisions are waiting for me in Player One?" and it answers from your installation. It can also start a chat with one of your agents, read what they are doing, and take up work they leave for it. This guide connects one tool with one address and one key. MCP explains what happens underneath.

Before you start

You need a tool that supports MCP servers, installed on the computer you work on: Claude Code, Codex, Cursor, VS Code, or another. If it runs on a different computer from your installation, your installation's address has to be reachable from there.

Keys are made on a page rather than by asking, on purpose. A key is shown once, to you, and none of your agents ever sees it.

1. Open the MCP page

Open Admin > Communications > MCP (/admin/mcp). At the top is the Server address, your installation's own address followed by /mcp, with a copy button.

You see the address, three numbered steps (Key, Harness and Watch), and at the foot a Signed connections section, which MCP covers.

2. Make a key

Under Key, open the dropdown and choose Create new key.

  1. Name it after the tool and the computer, such as "Claude Code (laptop)". The name is what the tool is called in your contacts.
  2. Choose what it may do. Read, Chat, Decide, Notify and Memory start on. Owner starts off, and MCP says when to turn it on.
  3. Choose an expiry: Never, 7, 30 or 90 days. The shortest one that suits is the safer one.
  4. Press Create key.

You see the key masked, with Show and Copy. It is filled into everything below it on the page. Copy it somewhere safe now: Player One keeps only a fingerprint of it, so it cannot be shown again. The page asks before you leave with a key you have not copied.

Each tool gets its own key, so you can cut one off without touching the rest.

3. Give the key to your tool

Under Harness, choose Claude Code, Codex, Cursor, VS Code or Other. Each has two ways to go on.

  • Let the agent do it gives you one prompt to paste into the tool. The tool adds Player One's server to its own settings, so the key is never saved in a project folder, then tells you what it needs for the server to load, then saves the Player One skill and says what it can do.
  • Do it myself gives you the command or settings for that tool, with your key filled in, and then a short prompt that holds no key.

A newly added server needs a nudge to load. The prompt says which, and it is usually one of these:

  • Claude Code: restart it, for example with claude --continue, and check that /mcp shows player-one connected.
  • Codex: restart it, for example with codex resume --last, and check that /mcp lists player-one.
  • Cursor: open Cursor Settings, check that player-one is switched on under MCP, and start a new chat.
  • VS Code: trust the player-one server when it asks.

You see the tool say what it can do in Player One, in a line or two.

A tool that runs only local servers, such as Claude Desktop, uses the Stdio-only clients panel at the foot of the page instead. It bridges to Player One and needs Node.js on that computer. The What a harness can do panel lists every action each permission opens.

4. Try it

Ask the tool something small. "List my Player One agents." "What decisions are waiting for me?" "Ask the research agent what we decided about pricing."

You see the answer in the tool. Open Manage > Contacts (/manage/contacts) and the tool is there as an agent of its own, named after its key. A chat it starts appears with your other chats, and its messages show as the tool's, never as yours.

5. Let your agents leave it work (optional)

Under Watch, copy the line and paste it into the tool when you want it to take up work Player One leaves for it:

"Connect to Player One and watch your inbox; take up what arrives and answer in the same chat."

Then ask one of your agents, "When the build fails, message my Claude Code with what failed." Player One never starts a tool or pushes to one, so the message waits until the tool looks. It works only while the tool is running. The tool answers in the same chat, and the agent that asked reads the reply there.

The same inbox is how your replies reach it. When the tool notifies you and waits, your answer arrives as a message it reads when it next looks, so tell it to check there.

Lost a key, or a laptop

Open Admin > Security > API keys on its MCP tab (/admin/api-keys/mcp). Each key shows its name, what it may do, when it expires and when it was last used.

  • Revoke cuts a key off at once and cannot be undone. The tool stays in your contacts, so its history stays attributed to it.
  • To keep the tool and change the key, choose it on the MCP page and press Reissue key, then confirm. The same key, with the same contact, permissions and history, gets a new secret, and the old one stops working everywhere. The tool needs the new prompt or settings before it can connect again.

If something goes wrong

Every call is refused. The key is revoked, expired, mistyped, or has been reissued since the tool was set up. Check its status on the MCP tab of API keys, then reissue it or make a new one.

The tool sees fewer actions than you expected. The key has fewer permissions. Reissuing does not change them, so make a new key with the ones you want.

The tool does not list player-one. It has not loaded the new server yet. Restart it, or follow the nudge in step 3.

It says it must slow down. A tool that asks too often is told to wait. It recovers by itself within a minute.

An agent's message never reaches it. The tool only finds messages when it looks, and only while it is running. Start it and paste the Watch line.

Where next