# matchcn

> A semantic index of shadcn-format UI components, exposed through a local stdio MCP server. Find components by description and retrieve their original registry install commands.

## Documentation

- [Full technical reference](https://matchcn.dev/llms-full.txt): complete plain-text documentation of installation, MCP parameters, result fields, matching dimensions, ranking, network behavior, benchmarks, and limitations.
- [Visible technical reference](https://matchcn.dev/#reference): the same documentation as readable HTML, with section anchors and source links.
- [Installation](https://matchcn.dev/#install): agent setup prompt and CLI configuration.

## Snapshot facts

Source version: 0.1.0, inspected September 20, 2026. Index: 6,119 components across ten registries (react-bits, magicui, aceternity, kokonutui, animate-ui, motion-primitives, shadcnblocks, shadcn-dashboard, assistant-ui, bundui). shadcn/ui is the format/installer ecosystem, not an additional indexed registry.

Tool: pick_component(brief, registry?, maxResults?). Outcomes: confident, shortlist, no_match. maxResults defaults to 3 and accepts 1–10 through MCP. Node.js 20+ required. Documented CLI: npx matchcn. The npm package is published and the source repository is public.

Ranking is deterministic for identical parsed inputs and tag records, but brief parsing and optional Resolve use classifier.dev over the network. Confidence measures matching/selection evidence, not general quality or benchmark accuracy. English briefs are more reliable in the current tests. No-match diagnostic candidates are not recommended results.

## Evidence

- [Full tagging run](https://matchcn.dev/reports/full-run.md): 6,119 components, 36,714 decisions combined across the original run (1,069 components, 6,414 decisions, approximately 2m43s tagging window) and the registry expansion run (5,050 components, 30,300 decisions, split across two IPs for the free daily quota).
- [Five-brief demo](https://matchcn.dev/reports/demo.md): raw results and caveats; not a labeled accuracy benchmark.
- [Pilot report](https://matchcn.dev/reference/pilot-report.md): tag confidence and criteria revisions.
- [Selection logic](https://matchcn.dev/reference/pick.ts.txt): response types and outcome rules.
- [Ranking](https://matchcn.dev/reference/match.ts.txt): local weighted distance calculation.
- [MCP adapter](https://matchcn.dev/reference/mcp-server.ts.txt): parameter validation and JSON text output.

## Project

- [Homepage](https://matchcn.dev)
- [Updates](https://x.com/matchcndev)
- [Builder](https://x.com/whosfranki)
- [Source license snapshot](https://matchcn.dev/reference/LICENSE.txt)

Independent MIT-licensed project. No affiliation or endorsement by shadcn, indexed registries, classifier.dev, or TypeSafe. Installed components retain their original publishers' license terms.