Use case · AI + MCP

A habit tracker your AI can actually use.

Habit Pocket hosts an MCP server, so Claude, ChatGPT, or any MCP-capable assistant can connect straight to your own habits — read your entries and stats, log values you dictate in plain words, run honest weekly reviews, test whether two habits actually move together, and create charts on your dashboard. Nothing to install, and your data is never sent to any AI company by us: you authorize your own assistant, and you can revoke it any time.

Get Started for Free

5 habits free. No credit card. Sync is free.

See it

Ask it, or just tell it.

Two conversations on a loop: one where the assistant digs through real entries to explain a bad week of sleep, and one where a messy week gets logged — and charted — in plain words. Hover to pause, or switch threads below the card.

Your AI assistant · connected to Habit Pocket

get_stats✓get_all_entries✓

Looked at your last 30 days. Sleep length is fine — 7.2 h on average. Three things do stand out: • Late coffee: on days with caffeine after 15:00, your sleep score averages 61 vs 78. • Bedtime swing: you range from 22:10 to 01:30 — the inconsistency tracks with tired mornings more than the average does. • Screens in bed: evenings you log screen time push bedtime back ~40 minutes. The late coffee is the strongest signal.

Log 7.5 hours of sleep, and mark meditation done.

log_entry✓log_entry✓

Done — Sleep 7.5 h and Meditation ✓ for today. That makes a 12-day meditation streak.

I was travelling Mon–Wed: ran on Tuesday, slept about 6 hours each night.

get_account_context✓log_entry ×4✓

Filled in: Run ✓ Tuesday, Sleep 6 h for Monday through Wednesday. Your run streak survived — Monday and Wednesday are excluded days for it.

Nice. Chart my sleep window, with sleep score on top.

create_chart✓

Created “Sleep window” — a candle chart from bedtime to wake-up with your sleep score overlaid. It’s on your dashboard.

Scenario · Catching up

The messy week, fixed in one message.

The most annoying moment in habit tracking is coming back after a gap. Describe the week like you'd tell a friend; the assistant turns it into entries — each one visible and fixable in the app afterwards.

I was travelling all week. Ran Tuesday and Thursday, slept badly the whole time — about 6 hours a night. No meditation. Fill it in.

get_account_context ✓list_habits ✓log_entry ×9 ✓

Filled in Mon–Sun: Run ✓ on Tuesday and Thursday, Sleep at 6 h for all seven nights, and Meditation left empty (it will read as missed). Your run streak survived — Tuesday and Thursday were your scheduled days.

Scenario · The Sunday review

A weekly review that reads your data, not your mood.

The server computes the same streaks and rates the app shows, and a built-in weekly_review prompt makes the whole ritual one click in Claude's menu.

Give me my weekly habit review.

get_account_context ✓get_stats ✓get_all_entries ✓

Solid week — 86% overall, your best since May. Held up: Meditation went 7-for-7 (streak now 19 days) and Sleep averaged 7.4 h, up half an hour. Slipped: Reading missed Thursday and Friday — both days you logged a late bedtime. One focus: your 10k step target hit on only 2 of 7 days but came close on 4 more; consider 8k, which your data says you can hold.

Scenario · The honest no

Sometimes the answer is “that's noise.”

An assistant reading real numbers can tell you when the pattern you hoped for isn't there — which is worth more than a chart that flatters you.

Does journaling improve my mood scores?

list_habits ✓get_all_entries ✓

Honestly: the data doesn’t support it yet. Mood averages 3.4 on journaling days vs 3.3 without — within noise for the 22 days you’ve tracked both. What does track with mood is sleep: 7+ hour nights average 3.9 the next day vs 2.8 after short nights. Worth re-checking journaling in a month when there’s more data.

Keep reading

FAQ

AI habit tracking — common questions

Claude (claude.ai, the mobile apps, Claude Desktop, and Claude Code), ChatGPT via its connectors, and editors like Cursor — anything that speaks MCP over HTTP. Claude and ChatGPT connect through a consent page with no key handling; other tools use an API key from Settings → Integrations.

No, never. Habit Pocket has no AI features of its own phoning home. Connecting via MCP is you authorizing your own assistant to read your own account — the conversation happens between you and the assistant you chose, and revoking access ends it immediately.

Yes — say “log 7.5 hours of sleep and mark meditation done” and it writes both entries, which sync to your phone like anything logged by hand. It can also backfill a messy week from a description, create habits, and add dashboard charts. It deliberately cannot edit, archive, or delete anything you built.

No. In claude.ai or ChatGPT you paste one URL into the connector settings and approve access in your browser. The REST API exists too, for people who do want to write code, but MCP is the no-code path.

The assistant reads the same server-computed stats the app shows — real streaks, rates, and values, not vibes. The built-in prompts explicitly tell it to say when a sample is too small or a pattern is noise, and to frame findings as hypotheses rather than conclusions.

Five habits. Thirty days.
See what shows up.

No credit card. Pick five habits. Come back in a month.