Add the DiceyTable MCP server to Claude Code

Run claude mcp add diceytable -- npx -y @diceytable/mcp-bridge@1 in a terminal, inside your project folder. From then on, Claude Code starts the DiceyTable bridge for you whenever it needs it.

The bridge is the small program that links Claude to a DiceyTable tab in your browser. It comes from npm, so there is nothing to download by hand. This page assumes you have saved a key with login, as Install the DiceyTable MCP bridge shows.

What you'll need#

• Claude Code, signed in. Run claude --version to check.
• Node.js 20.19 or newer. Run node --version to check. Claude Code uses the same npx you do.
• A saved key. Run npx -y @diceytable/mcp-bridge@1 status. It should name your account and your key. No key yet? Create an API key for your AI agent first.

Step 1: Add the server#

Open a terminal in your project folder and run:

claude mcp add diceytable -- npx -y @diceytable/mcp-bridge@1

Everything after -- is the command Claude Code runs to start the bridge. Claude Code prints an Added line when it has saved the entry.

Keep the name diceytable. Claude Code shows each tool as mcp__diceytable__<tool>, for example mcp__diceytable__session_connect. The instructions for your AI agent use that name too.

Should it work in one project or all of them?#

Claude Code calls this the scope. The command above uses the default, local, so DiceyTable only shows up in this one project.

To use DiceyTable in every project on this computer, add --scope user:

claude mcp add diceytable --scope user -- npx -y @diceytable/mcp-bridge@1

Both scopes live in your own ~/.claude.json file. That file belongs to you and is never part of a repository.

How do I share the setup with my team?#

Use the project scope. Commit a file named .mcp.json at the root of your repository:

{
  "mcpServers": {
    "diceytable": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@diceytable/mcp-bridge@1"]
    }
  }
}

This file holds no key, so it is safe to commit. Each person saves their own key with login. The first time someone opens the project, Claude Code asks them to approve the server.

Step 2: Check the connection#

Run:

claude mcp list

You should see diceytable with ✔ Connected. The very first check can say ✘ Failed to connect while npx downloads the bridge. Wait a moment and run it again.

Inside a Claude Code session, type /mcp to see the same status. Connected means Claude Code can start the bridge. Your key gets checked later, when the agent connects to DiceyTable.

Step 3: Connect a DiceyTable tab#

Ask Claude to "connect to DiceyTable". It replies with a pairing code. Type that code into a DiceyTable tab, as Pair a DiceyTable tab with your agent shows. Then check your agent is connected.

Local files and other folders#

The bridge only reads and writes local files inside allowed folders. Claude Code usually starts it in your project folder, and that folder is always allowed. To let the agent import art from somewhere else, add --allow-dir after the package name:

claude mcp add diceytable -- npx -y @diceytable/mcp-bridge@1 --allow-dir /path/to/card-art

Which local files an agent can read explains the rules.

Using an environment variable instead of login#

login is the easiest choice. If you would rather keep the key in an environment variable, add an env line to the diceytable entry in .mcp.json. It names the variable and never holds the key itself:

"env": { "DICEYTABLE_TOKEN": "${DICEYTABLE_TOKEN}" }

Claude Code fills in ${DICEYTABLE_TOKEN} from your shell when it starts the bridge.

If it doesn't work#

• Connection timed out at startup. The first npx download can take longer than Claude Code's 30-second limit. Start Claude Code with a longer one: MCP_TIMEOUT=60000 claude, or $env:MCP_TIMEOUT = "60000"; claude in PowerShell.
• The bridge won't start on Windows. Wrap the command: claude mcp add diceytable -- cmd /c npx -y @diceytable/mcp-bridge@1.
• You ran login after starting Claude Code. The bridge reads the key when it starts. Reconnect the server from /mcp.

For anything else, run npx -y @diceytable/mcp-bridge@1 doctor, then see Fix an agent that won't connect.

To remove the server later, run claude mcp remove diceytable. Disconnect, revoke or remove the agent covers the rest.

Next steps#

• Pair a DiceyTable tab with your agent: type the code and read the consent dialog.
• Check your agent is connected: three quick calls that prove it all works.
• Instructions to give your AI agent: a starting prompt and a CLAUDE.md snippet.