Help & FAQ

Go to the Install page, copy the prompt, and paste it into any Claude Code session. Claude will add the hook to your settings automatically.

The hook is a single entry in ~/.claude/settings.json that fires when a session ends and sends your usage data to your dashboard.

It can. Claude Code pipes session data to hooks via stdin, but only the first hook that reads stdin gets the data. If you have an existing hook that uses $(cat) or reads stdin, our hook needs a different approach.

The install prompt on our Install page handles this automatically — it tells Claude to detect existing hooks and use the right method.

If it still doesn't work, paste this into Claude Code:

Paste this into Claude Code:

Most common cause: No session has ended yet. The hook only fires when a Claude Code session ends — not while it's active.

Test it: Open a separate terminal and run:

This starts and immediately ends a session. Refresh your dashboard — you should see data.

Still nothing? Paste this into Claude Code to verify your hook is installed correctly:

The hook might be failing silently. Common causes:

1. The command got modified — especially -d @- at the end. It must be exactly -d @- (dash, at-sign, dash). Paste this into Claude Code to check:
   Show me the MyTokenTracker hook in ~/.claude/settings.json. Does the command end with exactly "-d @-"? If not, fix it.
2. Another hook consumes stdin first — see the "I already have Stop hooks" FAQ above.
3. Network issue — the hook runs curl which needs internet access. Test manually:
   curl -s https://mytokentracker.io/up

Token counts come directly from Claude Code's Stop hook event. If they look off, it's usually because:

• The session was very short (few tokens)
• Cache read tokens dominate — this is normal, often 95%+ of all tokens
• The model shown doesn't match what you expected — Claude Code may switch models

Use Quick Sync on the dashboard to calibrate accuracy against Claude's actual reported percentages.

Try logging out and back in. If the issue persists, try a hard refresh (Cmd+Shift+R on Mac, Ctrl+Shift+R on Windows).

If you still see a 500, open an issue on GitHub with what you were doing when it happened.

The install prompt works the same on Windows — just paste it into Claude Code. A few things to check:

1. Verify curl is available — it ships with Windows 10 and later. Open PowerShell and run curl --version. If you get a "not recognized" error, you may need to update Windows or install curl manually.
2. Settings file location — on native Windows, the settings file is at %USERPROFILE%\.claude\settings.json. In WSL, it's ~/.claude/settings.json inside the Linux filesystem. Make sure the hook was added to the right one for how you run Claude Code.
3. WSL users — if you run Claude Code inside WSL, everything works exactly like Linux. The paste-into-Claude-Code install method works identically.
4. PowerShell alias conflict — PowerShell aliases curl to Invoke-WebRequest by default. Claude Code hooks run via the system shell, not PowerShell, so this should not be an issue. If it is, ask Claude Code to use curl.exe instead of curl in the hook command.

If you're still stuck, paste this into Claude Code:

Several reasons this is expected:

• We only track Claude Code — Claude's limits include web, mobile, and API usage too
• Anthropic's formula isn't public — we estimate using published API pricing, but their internal calculation may differ slightly
• No syncs yet — the correction factor starts at 1.0x (no adjustment)

Fix it: Use Quick Sync on the dashboard. Enter Claude's reported percentages from claude.ai/settings/usage and we'll learn a correction factor. After 3-5 syncs, the numbers closely match.

No — this is the API-equivalent cost, not what your subscription charges. It's what you'd pay if you used Anthropic's pay-per-token API instead of your Claude subscription.

We use this because Anthropic's rate limits are cost-weighted — they use API-equivalent dollars to calculate your usage percentage. So this number drives the plan usage bars on your dashboard.

If your estimated cost is higher than your subscription price, your plan is saving you money (check the Plan Savings card in Data Insights).

Open Claude's usage screen and note your session % and weekly %. Enter them in the Quick Sync bar on your dashboard and hit Sync.

We calculate the ratio between Claude's number and ours (e.g., Claude: 34%, ours: 29% = 1.17x correction). This factor is stored and applied automatically. Each sync refines the factor — recent syncs count more than older ones.

Best time: When usage is 10-50% on Claude's screen. Very low or very high percentages amplify small errors.

How often: 3-5 syncs over different sessions is enough for the factor to converge. After that, occasional syncs keep it tuned.

Don't sync when: You have very little tracked data (under $0.10 of usage), or right after a fresh weekly reset.

Open Claude's usage screen. Under "Weekly limits" it shows something like "Resets Thu 10:00 AM". Set that same day and time in your Settings.

Session limit — a rolling 5-hour window. As older usage ages out, capacity frees up. This is the "Current session" bar in Claude.

Weekly limit — resets on a fixed day/time (e.g., Thursday 10 AM). All usage within the week counts against this budget. This is the "Weekly limits" section in Claude.

You can hit either limit independently. The session limit governs short-term burst usage; the weekly limit governs sustained usage over days.

No. The hook only sends aggregate usage statistics from Claude Code's Stop event. This includes token counts, model name, project directory name, and session duration.

Your prompts, Claude's responses, file contents, and any code you're working on are never transmitted. You can verify this yourself by inspecting the hook command in ~/.claude/settings.json — it only sends the JSON payload that Claude Code passes to Stop hooks.

All data is transmitted over HTTPS (TLS 1.2+) to mytokentracker.io. The hook uses curl with HTTPS by default — there is no plaintext HTTP fallback. Our SSL certificate is issued by Let's Encrypt and auto-renews.

Your usage data is stored in a MySQL database on our production server. Each user's data is isolated by their account — you can only access your own data when authenticated via GitHub OAuth.

We do not share, sell, or provide your data to any third parties. Your usage data is used solely to power your personal dashboard.

The hook is a single curl command that runs when a Claude Code session ends. It:

• Reads the session data that Claude Code pipes to it via stdin (token counts and metadata only)
• Sends that data to our API endpoint over HTTPS
• Does nothing else — no file access, no background processes, no persistent connections

The hook runs in the same sandbox as other Claude Code hooks. You can inspect the full command in ~/.claude/settings.json at any time, and remove it instantly by deleting the entry.

GitHub OAuth means we never store or handle your password. Authentication is delegated entirely to GitHub. We only receive your GitHub username, email, and avatar — the minimum needed to create your account.

This also means you benefit from any security features you've enabled on GitHub (2FA, passkeys, SSO) without us needing to implement them separately.

We receive the project directory name (e.g., "my-app") from the session data, which we use to group usage by project on your dashboard. We do not receive the full file path (e.g., /Users/you/code/my-app).

If your project directory name is sensitive, you can rename the folder or contact us to have specific project names scrubbed from your account.