Count lines of code without cloning the repository
You can count a GitHub repository's lines of code without cloning it: paste the URL into the OctoCounts web app and it downloads the source archive of the requested ref — not the git history — counts it with tokei, and returns files, total lines, code lines, comments, blanks, and per-language totals. Duration depends on repository size, upstream access, queueing, and deployment limits. The browser extension, report pages, .md twins, badges, and API all answer from the same archive-based pipeline, so no step ever requires a local clone.
Why avoid cloning just to count lines?
A full clone includes repository history as well as the current source tree. A source archive contains one revision without that history, which is enough for a line count. Transfer size and duration depend on the repository and network; shallow or partial clones can also reduce transfers. With OctoCounts, archive download and counting happen on the server.
How do I count lines without cloning?
- Open the OctoCounts web app and paste the repository URL — for example
https://github.com/react/react. - Optionally pick a branch, tag, or commit SHA. The analysis pins to the exact commit it resolves.
- OctoCounts downloads the source archive for that ref, runs tokei, and shows files, total lines, code, comments, blanks, and per-language rows.
- Share the report URL, export text/JSON/PNG, or add a live badge to your README.
If you keep reading code on github.com, the OctoCounts extension skips even the paste: it adds a SLOC card to the repository sidebar on every public repo page you visit.
Do git partial clones make no-clone counting unnecessary?
They narrow the gap without closing it. git clone --depth 1 truncates history but still fetches every file of the latest commit, and --filter=blob:none defers file contents until checkout — at which point running wc, cloc, or tokei fetches the whole tree anyway. A shallow partial clone of a monorepo can still be hundreds of megabytes on your disk. There are also read-only tricks that avoid git entirely — the GitHub API page explains why the API route still cannot give you lines — server-side archive counting is one way to avoid moving the source tree onto your machine.
Do the numbers match a count on a local clone?
Matching counts require the same actual source tree, tokei version, and effective counting rules, including extraction exclusions and repository attributes. A local checkout at the same commit may differ from an archive: export-ignore can omit paths and export-subst can change file contents. Record the full commit SHA, engine version, options, effective policy, and measurement date; retain the exported JSON and a saved-report link while available. Older reports may lack a complete policy. The filtering pilot measures option effects, not local-versus-server parity. See Git archive attribute semantics.
What is the fastest way to count lines without cloning?
For a one-off, open https://octocounts.com/github/:owner/:repo directly — a warm report can reuse cached counts because results are cached by repository, commit SHA, tokei version, and analysis options. For automation, the API's GET /api/reports/:reportId returns full JSON and the badge endpoints return a ready-made SVG. For scale: in OctoCounts' 2026 pilot, archive-based analysis of the react and vite repositories completed in 1.2–1.9 seconds of observed wall time per request; cache and queue state were not controlled.
When do you still need a clone?
Private repositories — OctoCounts analyzes public github.com repositories only and does not accept uploads — and any analysis that needs local tooling beyond line counts. In those cases clone (shallow if you like) and run tokei or cloc in the checkout; the full counting guide covers the clone-based commands too.
How can I verify a count without keeping a clone?
Export the report JSON and retain its full commit SHA, tokei version, generatedAt, selected options and effective countingPolicy. The ten-repository study archives complete reports rather than relying on a live cache. It measures filtering effects on GitHub source archives; it does not establish parity with a local checkout, and request wall time is not a cold-download benchmark.
Read the October 27, 2026 study and download its raw reports.