TokenEyez
Install free →

MCP setup

Ask Claude about your AI spend.

The TokenEyez MCP server plugs your usage dashboard into Claude Desktop, Cursor, and Claude Code so you can ask plain-language questions — "how much did I spend on Opus this week?" — without leaving your editor.

★ Pro feature
Before you start: you'll need Node.js 18+ installed (it comes with npm) and a TokenEyez Pro account. Total setup time on a fresh machine: about 90 seconds.

  1. Install the package

    Open a terminal — macOS / Linux: Terminal app · Windows: PowerShell — and run:

    npm install -g @tokeneyez/mcp

    This installs the tokeneyez CLI and the MCP server globally on your machine.

  2. Sign in

    Same account as the Chrome extension — you only do this once per machine.

    tokeneyez login

    A browser tab opens. Click Authorize. Done — you'll see Signed in as you@example.com in the terminal.

  3. Add to your AI client

    Pick the one you use. You can repeat for multiple clients.

    One command — installs a Stop hook so every Claude Code turn lands in your dashboard automatically:

    tokeneyez install

    After your next Claude Code turn, run tokeneyez today to see it land. Remove anytime with tokeneyez uninstall.

    What it does: writes a single line into your Claude Code settings.json — a Stop hook that fires tokeneyez-hook after every turn. No prompt content leaves your machine; only token counts and model name.

    Open your Claude Desktop config file:

    • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
    • Windows: %APPDATA%\Claude\claude_desktop_config.json
    • Linux: ~/.config/Claude/claude_desktop_config.json

    Add a mcpServers entry (or merge into your existing one):

    { "mcpServers": { "tokeneyez": { "command": "tokeneyez-mcp" } } }

    Quit Claude Desktop completely (Cmd-Q / Alt-F4) and reopen. The MCP icon (a tiny hammer) appears in the bottom of the chat input — that confirms tokeneyez is loaded.

    If the file already has other MCP servers, just add "tokeneyez": { ... } as a new key inside the existing mcpServers object. Don't replace the whole file.

    Cursor's MCP integration ships out of the box. One command does it:

    tokeneyez install-cursor

    Restart Cursor. Open the AI chat → ⚙ settings → MCP servers → tokeneyez should be listed and green.

  4. Try it

    Ask any of these in your AI client (or invent your own):

    • "How much did I spend on AI this week?"
    • "Which model burned the most tokens today?"
    • "What's my Pro quota status?"
    • "Show me my top 5 most expensive prompts this month."
    • "Am I close to my monthly budget?"

    The MCP answers in plain language and shows live numbers pulled straight from your dashboard.

Troubleshooting

"npm: command not found"

Node.js isn't installed yet. Grab the LTS installer from nodejs.org. After installing, close and reopen your terminal, then retry step 1.

"EACCES: permission denied" on npm install

This is an npm permissions issue on macOS / Linux. The cleanest fix is to use nvm to manage Node, which puts npm under your home directory and skips the permissions trap.

Quick workaround: prefix with sudo — sudo npm install -g @tokeneyez/mcp.

I ran tokeneyez login but it says "browser didn't open"

The CLI prints a URL and a token. Copy the URL into any browser, sign in, and you'll see an Authorize button. Once you click it, the CLI completes automatically.

Claude Desktop doesn't show the hammer / tokeneyez isn't listed

Three things to check:

"Not signed in" when the MCP runs

Run tokeneyez login again. Sessions can expire if you've signed in from another device. The browser dance is the same — about 5 seconds.

Can I remove the MCP later?

Yes. Run tokeneyez uninstall for the Claude Code hook, or tokeneyez uninstall-cursor for Cursor. For Claude Desktop, remove the "tokeneyez" entry from your mcpServers in claude_desktop_config.json and restart.

To remove the package entirely: npm uninstall -g @tokeneyez/mcp.

What data does the MCP send?

Only what's already in your dashboard — daily totals, per-model breakdowns, quota state. Your prompts and responses never leave your machine. The MCP server runs locally and reads from the TokenEyez API using your session token, which lives in ~/.tokeneyez/session.json.