jscpd integrates into AI-powered development workflows through three complementary mechanisms: the AI reporter, agent skills, and an MCP server.
The ai reporter produces compact, token-efficient output designed to be piped directly into an LLM prompt or agentic pipeline. It uses common-path-prefix compression and omits code fragments and colors — just the clone locations and a summary.
jscpd --reporters ai /path/to/sourcecpd --reporters ai /path/to/sourcesrc/utils/ auth.ts:10-25 ~ helpers.ts:40-55
src/utils/auth.ts 30-45 ~ 80-95
src/ utils/auth.ts:10-25 ~ api/routes.ts:5-20
---
23 clones · 4.2% duplication
Benchmarked on the fixtures/ directory (212 clones, 347 files):
| Reporter | Output size | Estimated tokens |
|---|---|---|
console (default) |
~21,800 chars | ~5,400 |
ai |
~4,500 chars | ~1,100 |
~79% fewer tokens than the default console reporter.
Add --summary for a compact refactoring-hotspot overview — top files and folders by tokens, lines, size, and a complexity estimate. In the ai reporter each entry is one line with all metrics inline, so an agent gets the full picture for a handful of tokens:
Summary by tokens (321 files, 129 folders):
files (tokens/lines/size/cx/dup%):
src/core/files.ts 2052/363/11.4K/80/0.0%
...
folders (files/tokens/lines/size):
src/core 8/5264/843/26.5K
...
cpd --reporters ai --summary --no-tips /path/to/sourceSee rust.md for the metric definitions and --summary-top / --summary-by options.
jscpd ships two AI agent skills that teach coding assistants how to use jscpd and refactor detected duplications.
Covers all CLI options, the AI reporter output format, and configuration file syntax.
npx skills add kucherenko/jscpd --skill jscpdA guided process for reading clone output, choosing the right extraction strategy, applying the refactor, and verifying the clone is eliminated.
npx skills add kucherenko/jscpd --skill dry-refactoringAfter installation, ask your agent to "find and fix code duplication" and it will invoke jscpd with the right options and act on the results.
jscpd speaks the Model Context Protocol (MCP) over two transports, exposing detection capabilities as tools that AI assistants can call directly from the editor. Start a server once against your codebase, then let your AI assistant check any snippet for duplication on demand — no CLI invocation needed.
| Transport | Command | Engine |
|---|---|---|
| stdio | cpd --mcp /path/to/project (or jscpd --mcp) |
Rust v5 |
| Streamable HTTP | jscpd-server /path/to/project |
Node.js v4 |
The cpd/jscpd v5 binary serves MCP over stdio directly — the transport most MCP clients spawn-and-manage themselves, with no port, no network policy, and the Rust engine's scan speed. The project is scanned once at startup (log line on stderr); snippet checks run against in-memory token hashes, so they answer in milliseconds even on large codebases.
cpd --mcp /path/to/project
# All detection options apply to the scan and to snippet checks:
cpd --mcp --min-tokens 30 --format javascript,typescript /path/to/projectClient configuration (Claude Desktop, Claude Code, Cursor, APM, ...):
{
"mcpServers": {
"jscpd": {
"command": "cpd",
"args": ["--mcp", "/path/to/project"]
}
}
}The server implements MCP protocol revision 2025-06-18 (also accepting 2025-03-26 and 2024-11-05 clients) and exposes four tools:
check_duplication(code, format, limit?)— check a snippet against the scanned project;formataccepts format names (javascript) or file extensions (js)get_file_clones(path, limit?)— clones involving one file, for file-scoped refactoring;pathis scan-root-relative (as shown in results) or absoluteget_statistics()— totals and per-format statistics from the last scancheck_current_directory(limit?)— re-scan the configured paths and return updated counts plus the clone list
Tool results are compact JSON in a text content block. Every clone/match list is sorted biggest-first (by tokens) and capped by the optional limit argument (default 100) — the accompanying clones/count field always reports the untruncated total, and truncation is flagged with a note.
jscpd-server serves MCP over HTTP plus a REST API — use it when several clients share one long-lived server or you need the LevelDB store.
npm install -g jscpd-serverStart the server:
jscpd-server /path/to/projectOptions:
--port— Port number (default: 3000)--host— Host to bind (default: 127.0.0.1)--allowed-origin— ExtraOriginhostname accepted by the MCP and REST endpoints (repeatable)--allowed-host—Hosthostname the MCP and REST endpoints answer on (repeatable)--store leveldb— Use LevelDB persistent storage- Plus all standard jscpd detection options
Add to your MCP client config (e.g. Claude Desktop):
{
"mcpServers": {
"jscpd": {
"type": "streamable-http",
"url": "http://localhost:3000/mcp"
}
}
}The endpoint serves protocol revision 2026-07-28: requests are direct and stateless, carrying their protocol version and client capabilities in the per-request _meta envelope, so there is no initialize handshake and no Mcp-Session-Id. Clients discover the server with server/discover. 2025-era clients keep working through the SDK's stateless legacy fallback.
/mcp, POST /api/check, POST /api/recheck, and GET /api/stats validate the Origin and Host headers, as the transport specification requires. Loopback origins and hosts are allowed by default; add --allowed-origin for a browser client served under another name, and --allowed-host to pin extra hostnames a reachable deployment answers on. A concrete --host is always included in the Host allowlist.
| Method | Path | Description |
|---|---|---|
POST |
/api/check |
Check a code snippet for duplications. Body: {"code": "...", "format": "javascript"} |
POST |
/api/recheck |
Trigger a re-scan of the directory |
GET |
/api/stats |
Get overall project duplication statistics |
GET |
/api/health |
Health check — returns { status, workingDirectory, lastScanTime } |
GET |
/ |
API info with endpoint listing |
Available MCP tools exposed via the /mcp endpoint:
check_duplication— Check a code snippet for duplications (inputs:code,format)get_statistics— Get project stats (no inputs)check_current_directory— Re-scan the working directory (no inputs)
Snippet checking uses an ephemeral in-memory store per request for isolation — no cross-request contamination, automatic cleanup, concurrent-request safe.