bliptracker is the scope for your AI chats — a Chrome extension that labels your claude.ai conversations with a 🔴 / ✅ prefix so you can see at a glance which ones are waiting on you versus done — on desktop and on mobile. → bliptracker.xyz
⚠️ Unofficial community tool. Not affiliated with or endorsed by Anthropic. It uses undocumented claude.ai endpoints and may break without notice.
➕ Add to Chrome — the easy way.
Or load from source (unpacked)
- Clone or download this repo.
- Go to
chrome://extensions→ enable Developer mode → Load unpacked → select theextension/folder. - A setup tab opens automatically — copy the preference prompt into your Claude profile settings (it links you straight there).
After installing, do the one-time setup (below) and you're done.
| Waiting-on-you, labelled | The queue popup | One-time setup |
|---|---|---|
 |
 |
 |
- A one-time preference prompt in your Claude profile makes Claude end
substantive replies with a machine-readable marker:
🔴 Waiting on you: <reason>or✅ Resolved — safe to archive. This runs everywhere you use Claude, including mobile. - The extension reads that marker and renames the chat title with a
🔴/✅prefix. Because titles are stored server-side, the labels sync to the claude.ai mobile app automatically — you only run the extension on desktop. - Click the toolbar icon for a “waiting on you” queue of your 🔴 chats; the badge shows the count.
- Not ready to deal with a chat yet? Snooze it from the popup — six presets
(1 hour, this evening, tomorrow 9am, 3 days, next Monday, 1 week) or a
specific date and time, picked on a half-hour grid. It leaves the queue and
the badge, its title becomes
💤🔴 …(synced to mobile like any label), and it sits in a collapsed Snoozed section until it wakes — automatically, via a one-shot alarm, even between sweeps. Wake times live in local extension storage, so snooze schedules are per-machine.
Classification is deterministic — a string match on the marker Claude wrote, never an after-the-fact guess.
Right-click the toolbar icon → Options (or the Settings link in the popup):
- Enabled — master switch.
- Mirror status to stars — also star 🔴 chats / unstar ✅ ones, so the Starred section is your mobile queue. Off by default so it doesn't collide with stars you set manually.
- Sweep interval — how often to re-check all chats. The chat you're actively viewing updates instantly regardless.
Three layers, each with one job:
- Orchestrator (
extension/lib/orchestrator.js) — platform-agnostic. Owns the sweep, the on-demand re-check, theseencache and the badge. Knows only the adapter interface, never an endpoint. - Service adapters (
extension/adapters/*.js) — one per platform, running in the service worker. Own endpoints, auth and response parsing, and normalise everything to{id, name, isStarred, updatedAt, lastAssistantText, isTemporary}. Registered inadapters/index.js. - DOM adapters (the
DOM_ADAPTERSmap inextension/content.js) — one entry per host, running in the page. Own selectors:currentId(location)andtitleNodes(id).
The split follows a process boundary, not just a file boundary: service
adapters need the worker's cookie context, DOM adapters need the page. That's
why content.js carries its own map rather than importing the registry.
Beyond the layering:
- Two triggers, one brain.
content.js(instant, the chat you're viewing) and the alarm sweep (periodic, all chats incl. mobile-finished) both funnel throughcheckConversation/applyStatusin the orchestrator. No classify or rename logic is duplicated in the content script. - Idempotent renames. Renaming bumps a chat's
updated_at, so the next sweep re-checks it —titleTransform()is idempotent, so the re-check is a no-op. Convergence by design, no bookkeeping. - Best-effort DOM paint. The instant sidebar update writes to the page's DOM; if React reverts it, the resulting mutation re-triggers our observer and we re-apply. The server-side rename is the source of truth regardless.
- No backend. All requests go directly to the platform with your own session; nothing is sent anywhere else. See PRIVACY.md.
Used only by the Claude service adapter (extension/adapters/claude.js); no
other module references them. All under /api/organizations/{orgId}:
| Method | Path | Use |
|---|---|---|
| GET | /chat_conversations?limit=N&offset=M |
list, updated_at desc (incl. is_starred); a short page means end-of-list |
| GET | /chat_conversations/{uuid} |
full chat incl. chat_messages[] |
| PUT | /chat_conversations/{uuid} {"name":…} |
rename |
| PUT | /chat_conversations/{uuid} {"is_starred":bool} |
star |
| DELETE | /chat_conversations/{uuid} |
delete (unused) |
See ROADMAP.md for shipped features, what's next, and planned future work (e.g. a popup "Resolve" button, one-click preference injection, bulk-archive, and multi-AI adapters).
Pure JS, no build step, no dependencies. Load extension/ unpacked at
chrome://extensions.
Tests use Node's built-in runner (Node ≥ 20):
npm test # node --test 'tests/**/*.test.mjs'tests/helpers/ holds a ~50-line chrome stub and an in-memory fake adapter,
so the orchestrator can be driven without a browser. Adding a platform means
writing a service adapter and a DOM adapter — see
ROADMAP.md.
Found a bug, or got an idea? Open an issue — feedback on the onboarding especially welcome. PRs are warmly received; it's a small, no-build codebase, so jump straight in.
MIT © 2026 Andrew Miller