It reads the answer to a request the page was going to make anyway. It doesn't forge API calls, doesn't scrape rendered HTML. X makes the authenticated request; the extension tees the JSON response.
A Chrome extension that turns your already-logged-in X tab into a local export tool. When you search or scroll, X's own JavaScript calls its internal GraphQL backend and renders the JSON; the extension reads those response bodies in the page and exports them to JSONL, CSV, or markdown. You never authenticate to anything — it runs inside the tab you already signed into.
Reading X's own response — instead of forging requests or parsing rendered HTML — is the one property that separates it from a DOM scraper:
| Compared to | They do | This does instead |
|---|---|---|
| DOM scrapers / Nitter | Parse rendered HTML; race the paint; break on UI redesigns; run on a third-party server | Reads the structured JSON X itself fetched — captures followers / blue / views / quotes the rendered page omits; survives UI redesigns; stays inside your own session |
X is a single-page app — open a search or profile and it runs a GraphQL query
(SearchTimeline, UserTweets, ListLatestTweetsTimeline, HomeTimeline, …) and
renders the JSON. The extension reads the answer to a request the page makes anyway:
X's JS fires SearchTimeline ─► patched fetch/XHR tees the response ─► parse ─► JSONL / CSV / markdown
inject.js(page main world) patcheswindow.fetch+XMLHttpRequestto tee the GraphQL response bodies — MV3 removed blocking-webRequestbody access, so this is how you read them. It never forges a request; X makes the authenticated call, the extension reads the result.content.js(isolated world) parses X's timeline JSON into flat records (parser parity-tested byte-for-byte against a reference implementation), dedupes by tweet id, and exports.
Per record: id, handle, name, text, time, url, likes, replies, reposts, followers, blue, verified, views, quotes, lang, is_retweet, source.
chrome://extensions→ enable Developer mode- Load unpacked → select this folder
- Open x.com logged in — an API-God panel appears bottom-right
- Search or scroll; the counter climbs as responses land. Preview shows what's captured so far. Save with JSONL, CSV, or Markdown, or Copy the JSONL to the clipboard.
Flip Autoscroll: ON and the panel jumps to the bottom of the page about once a second so X keeps loading the next page — each hit pulls in a whole page of posts. The extension still fetches nothing itself. Set a minutes value next to the toggle to stop after a fixed time (a live countdown shows in the panel); leave it blank to run until the timeline runs dry. Either way it stops on its own once no new posts arrive. Leaving it running is automated pagination; whether that fits your use of X is the operator's call (see Scope).
Preview opens a live pane of what you've captured. Type in its box to filter the
captured set (matches handle, name, or text); press Enter or Go X to run that text
as a search on X instead. Running it on X is a full navigation on purpose — it reloads the
page so inject.js re-patches fetch at document_start before X's bundle grabs its own
reference, which is the cleanest capture path.
Parlay runs a chain of searches back-to-back, each for a duration you set — e.g.
cats for 1 min, then dogs for 3 min, then birds for 5 min. Add legs (term +
minutes), press Start parlay, and it navigates to each search in turn, autoscrolls for
that leg's minutes, and moves on; every leg's posts accumulate into one deduped set you
export at the end. Because each leg reloads the page, the plan and the captured records are
parked in the page's own sessionStorage (a standard web API, x.com-scoped, local only, no
extra permission) so they survive the reloads; the run is discarded on Stop parlay,
Clear, or closing the tab. Like autoscroll, a parlay is automated pagination — the
operator's call (see Scope).
Web Store listing assets (1280×800 screenshots + 128×128 icon) live in store/.
Runs under your own account and session, over your own view — the same posts the page already showed you. No firehose, no full-archive, no key: it's bounded to what your session can see, at browsing speed. Automated/bulk access to X's internal endpoints can violate X's terms and risk the account; that call is the operator's. Capture and export only — it makes no network calls of its own.
MIT © 2026