Copy/Paste Detector for Source Code
npx jscpd .Built for codebases where agents write code
jscpd . --threshold 3Runs in pre-commit and CI. "An agent will argue a comment, but a failing build just stops it." — @nark3d
Set up the gate →jscpd --mcpMCP server built into the binary. Claude, Cursor or any MCP client can check for clones before writing more.
MCP server docs →npx skills add kucherenko/jscpdSkills that run the whole workflow: find duplicates, extract them, remove dead code, simplify complex files, then check the numbers improved.
Install the agent skill →jscpd . --reporters aiAbout 79% fewer tokens than the default reporter. Pipe it straight into your agent loop.
See the token benchmark →jscpd . --dashboardOne screen: a health score on top, then duplication, the most complex files and dead code. Fix the worst first.
Project dashboard →jscpd . --dead-codeUnused files, exports and imports in JavaScript, TypeScript and Python — and in Rust, read from the compiler's own diagnostics — each with a confidence score. --complexity ranks the most complex files without running clone detection.
jscpd . --history v5.0.0..HEADOne scan per commit, shown as a chart and a table in your terminal. See whether agents add duplication faster than you remove it, and how far you can tighten the threshold.
Duplication trend over history →Finds the copies that don't look like copies
jscpd .Type-1: identical tokens. Whitespace, layout and comments don't count. Every run reports these, starting from --min-tokens 50 and --min-lines 5.
jscpd . --ignore-identifiers --ignore-literalsType-2: the same code with different variable names or literal values. --ignore-annotations also drops @Decorator lines. Keywords keep their meaning, so return never matches retry. Reported as renamed.
jscpd . --max-gap-lines 2 --similarity 0.7Type-3: a copy with a few inserted or changed lines (--max-gap-lines N), or a JavaScript/TypeScript function with the same structure (--similarity RATIO, compared by syntax tree). Reported as similar with a score.
How teams wire it
- Pre-commit
- CI gate
- Agent loop
- Agent skill
Block the commit before the clone lands.
npx jscpd --threshold 3 .Fail the pull request when duplication crosses the line.
# .github/workflows/jscpd.yml
- uses: kucherenko/jscpd@v5
with:
threshold: 3
Give the agent the same detector, over MCP.
// MCP client config
{ "mcpServers": { "jscpd": {
"command": "jscpd", "args": ["--mcp", "."] } } }
Teach the agent the workflow, not just the tool: scan, refactor duplicates, remove dead code, simplify complex files, check the health score improved.
npx skills add kucherenko/jscpdWho Uses jscpd
10M+ npm downloads / month · ~5,000 repositories declare jscpd on GitHub's dependents graph · 15,500+ workflow files reference Super Linter, which runs jscpd by default
Core
A decade refining duplicate detection, now rewritten in Rust for native speed.
How detection works →JavaScript, Python, Java, Go, Rust, C++, TypeScript, Ruby... If you can write it, we can scan it.
View supported formats →A Rust engine with parallel detection across format groups. A 159 MB codebase takes 3.4 seconds. No Node.js runtime, just one native binary.
See benchmarks →One command, everywhere: your laptop, CI/CD, that ancient Jenkins server nobody wants to touch.
Get started →HTML, JSON, XML and README badges. Make technical debt visible (and slightly embarrassing).
Explore reporters →Set a threshold and fail the build, or commit a baseline and fail only on clones added since.
Configure thresholds →Vue, Svelte, Astro and Markdown files are tokenized per block, so a <script> in a .vue file can match a .ts file. --cross-formats "js-ts" compares related formats in one pool, catching clones between .js and .ts.
--blame shows who wrote each clone, from git history, so you know who to ask before you refactor.
One of the most trusted tools in the ecosystem. Join the developers who rely on jscpd every day.
View on npm →54 papers, theses and patents use or cite jscpd: a measurement instrument in ACM TOSEM, two ICLR papers and an 832-million-line industrial dataset, a baseline for new detectors, a component of five patented systems.
jscpd in research →54 verified blog posts, talks and case studies: real configs and numbers from teams wiring jscpd into CI gates, pre-commit hooks and AI agent loops.
jscpd in articles →Integrations
uses: kucherenko/jscpd@v5 scans the repository, fails on threshold and uploads SARIF to Code Scanning. No install step.
Works with the pre-commit framework, Husky or a plain shell hook in .git/hooks. Blocks the commit when duplication exceeds the threshold.
--reporters sarif writes a SARIF file that GitHub Code Scanning renders as inline alerts on the pull request.
The codeclimate reporter feeds GitLab Code Quality and openmetrics feeds its metrics reports, so duplication shows on the merge request.
Embed detection in your own tools with the Rust crates (cpd-finder, cpd-core), or run the CLI and parse its JSON from any language.
How much copy/paste ships in GitHub's trending repos? A daily pipeline runs jscpd on each and publishes the results.
See today's results →Fresh from GitHub Trending
Developers Are Talking
One of the most important tools you can add to CI when building with AI: github.com/kucherenko/jscpd
へぇ 「jscpd v5 は Rust 製で速い。 26.7 万行 / 127ms。CI コストは実質ゼロ」
I highly recommend jscpd: github.com/kucherenko/jscpd Can’t live without this in my projects now
TIL: jscpd
npmjs.com/package/jscpd がDartコードのコピペ検出も対応してることを dart-lang/sdk の issue で知って試したら普通に動いた( ´･‿･｀) 利用例: `jscpd . --pattern "**.dart" --min-lines 20`
One of my favorite tools during a code audit is jscpd. It detects copy/pasted code. In a codebase I’m currently reviewing, nearly half of the JS is duplicated! jscpd lists the specific duplicated spots so we can consider a cleanup.
Get AI to reduce duplicated code, reduce the jscpd number
I du regular sweeps with jscpd, ran a Socrates analysis yesterday and it was in the single digits. Could purge more, but writing overly generic code has its own problems.
Same here, jscpd and the linter both sit in our CI gate. The thing that surprised me is it holds up better with agents than a review comment does. An agent will argue a comment, but a failing build just stops it, so the duplication never gets in.
jscpd というコピペコードを検出する静的解析ツールいいかもしれない 結局コード品質を担保するにはAIに逐一指示するというアプローチよりは静的解析ツールとかでガチガチに縛ってしまった方が気持ち的に大胆にリファクタリングしやすい
just added a jscpd pre-commit hook and why did i not do this ages ago
Problem: You want to find copy/pasted code. Solution: jscpd. It finds copy/pasted code in over 150 languages. I just ran this command on a JS project: npx jscpd --ignore "**/node_modules/**" It found over 6,000+ lines of duplicated code.
I break my specs into directed acyclic graphs in order to identify what work depends on other work and which tasks can be worked in parallel. I also run jscpd in pre-commit which helps cleanup duplicate code and interfaces. See my GitHub DAG project as an example of this
a good way to minimize slop in a typescript codebase is a forced validation command of knip, jscpd and linter of your choice. the more static validation you have the better.
重複コードを消すツール！ これも良さそう！！
Been experimenting with jscpd to spot similar code that I might be able to refactor, with the goal of simplifying and optimising and hopefully shave some bytes off my apps. Works with 150+ formats.
I’ve got jscpd in the pre-commit hook for the same reason. It catches the duplicate before commit instead of in review, and review misses copy-paste because each diff reads fine alone. Gets more useful the more an agent repeats the same helper.
知らせる層は knip(デッドコード)と jscpd(重複)。 こっちは止めずに、気づかせるだけの設定 あとは"入れて終わり"にしたくなかったので、既存のコードで溜まってた借金を返すEpicも立てて、子Issue10件で完走
jscpd is genuinely one of those quietly-everywhere dev tools, congrats on the milestone. a small OSS project really can outlast flashier ones…
Problem: Your code base probably contains a lot of copy/pasted code, but it’s hard to find. Solution: jscpd jscpd is a command line tool that finds copy/pasted code and reports specific duplicated lines. Works for over 150 languages.
Contributors
Supported by
