Spec Driven Development
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
제작자가 쓴 설명(영어 원문)
Deterministic visual inspection for AI agents. Overlap, overflow, contrast, quality.
이 세 가지는 절대 하나의 점수로 합치지 않아요. 의견, 측정, 인기는 서로 다른 것이니까요.
이 서버는 아직 쉬운 말로 된 요약을 쓰지 못했어요. 아래는 레지스트리에서 읽어 온 사실이에요.
쓰고 있는 AI 앱을 고른 다음 설정을 복사하세요.
제작자가 레지스트리에 설치 정보를 올리지 않았어요. 아래 제작자의 README를 확인하세요.
내 컴퓨터에서 실행돼요. 설치하면 AI 앱에 도구 목록이 나타나요.
누구나 볼 수 있는 정보를 점검했고, 코드는 읽지 않았어요. 해당하지 않는 항목은 개수에서 뺐어요. · 2026년 11월 8일 점검
레지스트리 이름이 GitHub 계정 AbduljabbarBXR 기준으로 확인되었고, 이 계정이 저장소를 소유하고 있어요
아직 GitHub에서 이 저장소를 읽어 오지 않았어요
아직 GitHub에서 이 저장소를 읽어 오지 않았어요
내 컴퓨터에서 실행되는 서버라 인터넷 엔드포인트가 없어요
내 컴퓨터에서 실행되며, 대신 로컬 설정을 사용해요
내 컴퓨터에서 실행돼요. 도구 목록은 인터넷 엔드포인트가 있는 서버에서만 읽을 수 있어요
레지스트리에서 이 카테고리의 활성 서버 전체와 비교했어요. 스타는 게시자가 저장소를 소유한 경우에만 세요.
해당하는 기본 점검 1개 중 1개를 통과했어요. 코드는 읽지 않았어요. 키나 토큰을 넣기 전에 서버가 어떤 권한을 요구하는지 확인하고, 꼭 필요한 권한만 있는 키를 쓰세요.
레지스트리에 필수 키가 나와 있지 않아요. 다만 처음 쓸 때 로그인을 요구하는 서버도 있어요.
내 컴퓨터에서 실행돼요. AI 앱이 필요할 때 실행하고, 데이터는 제작자의 서버를 거치지 않아요(서버 자체가 온라인 서비스를 호출하는 경우는 제외).
GitHub 저장소는 마지막으로 – 변경됐고, 최신 버전(0.2.2)은 2개월 전 레지스트리에 올라왔어요.
Claude Desktop, Claude Code, Cursor, VS Code처럼 MCP를 지원하는 앱에서 쓸 수 있어요. 내 컴퓨터에 Rust도 있어야 해요.
같은 카테고리에는: Spec Driven Development, Remote Desktop Commander, Commonly
같은 카테고리에서 다른 제작자가 만든 서버예요. 스타 순으로 정렬했어요.
카드 보는 법:↓ 주간 다운로드i★ GitHub 스타i세 기둥: 리뷰 · 점검 · 사용 정도i업데이트 = 최신 버전을 레지스트리에 공개한 날i
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.
Read Commonly pods and post as a Commonly agent you create, with a token from your Commonly account.
제작자의 저장소에서 쓰인 그대로 가져왔어요. PopNx MCP가 검토하거나 보증한 내용이 아니에요. HTML, 이미지, 스크립트는 제거했어요. 링크를 누르면 이 사이트를 벗어나요.
Deterministic visual inspection for AI agents. The eyes a coding agent never had.
Cornea gives a coding agent that builds web pages a real way to *see* them. As a token-cheap, deterministic structural model it can reason over exactly, instead of megabytes of screenshots and raw DOM dumped into context.
An agent doesn't need a *photo* of a page. It needs to know is my layout broken, and how. Cornea computes an abstract *visual geometry model* of every element (box, position, z-order, computed styles) and derives inspection conclusions. overlap, overflow, contrast, quality. Then exposes them as native tools on three surfaces: CLI, MCP, and HTTP API.
One promise holds everything together: same input → byte-identical output. No Chromium, no sub-pixel variance, no server. A single ~1.4 MB binary.
# Build (Rust 1.98+, edition 2024)
cargo build --release
./target/release/cornea tests/fixtures/sample-bugs.html 360
Requires only
cargo. No browser, no Node, no system dependencies. Builds fine on a phone-class device.
Or install the `cornea` command from npm (no Rust toolchain needed):
npm install -g optrex # provides the cornea command
cornea --help
Take this page below. It looks fine as source. But it's hiding four layout bugs. Cornea finds every one.
<!-- tests/fixtures/sample-bugs.html (excerpt) -->
Card 1 Card 2 Card 3
Left overlap Right overlap ... Hard to read on white
Run it and Cornea reports the damage instantly:
$ ./target/release/cornea tests/fixtures/sample-bugs.html 360
{ "html_file": "tests/fixtures/sample-bugs.html", "viewport_w": 360.0, "element_count": 17, "est_tokens": 1068, // <-- entire page read for ~1k tokens "report": { "total_elements": 17, "visible_elements": 14, "overlaps": [ ... 7 collisions ... ], "overflows": [ ... 1 clipped ... ], "contrast": [ ... 2 AA failures ... ], "quality": { "score": 0.06, "label": "broken" } } }
Each finding is precise and actionable:
| Finding | Detail |
|---|---|
| Overlap | section.row ⇄ div.overlap-right, area 20000 px². The absolutely-positioned boxes cover the cards |
| Overflow | div.overflow-bad: right edge 600 exceeds viewport 360 → clipped |
| Contrast | black on blue: ratio 2.44:1. Fails WCAG AA (needs 4.5) |
| Contrast | #cccccc on white: ratio 1.61:1. Fails WCAG AA |
That is the value: hundreds of tokens, not hundreds of kilobytes, and a deterministic answer the agent can act on and re-verify.
Cornea is one engine, three doors. All three returns identical inspection JSON because they funnel through a single shared dispatch.
cornea <file.html | http(s)://url> [viewport_width] [viewport_height] [--js]
$ cornea page.html 360
{ "html_file": "page.html", "viewport_w": 360.0, "element_count": 42, "json_bytes": 8124, "est_tokens": 2193, "report": { "total_elements": 42, "visible_elements": 38, "overlaps": [], "overflows": [], "contrast": [], "quality": { "score": 1.0, "label": "good" } } }
cornea --serve
An agent calls layout.* tools directly; it passes the page source, or a live URL (see URL capture below), per call:
→ {"method":"tools/call","params":{"name":"layout.overlaps",
"arguments":{"html":" A ..." ,"width":360}}} ← {"id":1,"result":{"content":[{"text":"[{\"a_sel\":\"...div \u21c4 ...div\",\"area\":20000}]"}]}}
| Tool | Returns |
|---|---|
layout.inspect | Full visual model + report |
layout.overlaps | Elements whose boxes collide |
layout.overflow | Clipped / collapsed / off-screen |
layout.contrast | WCAG AA ratios for text elements |
layout.quality | 0..1 health score + issue list |
layout.fidelity | Which CSS features are exact vs approximated |
cornea --serve-http [addr] # default 127.0.0.1:8080
curl -s -X POST http://127.0.0.1:8080/inspect \
-H 'Content-Type: application/json' \ -d '{"html":" bady ","width":360}'
{"total_elements":1,"contrast":[{"selector":"...>p","fg":"#cccccc","bg":"#ffffff",
"ratio":1.61,"pass_aa":false}]}
| Route | Method | Returns |
|---|---|---|
/inspect | POST | Full report |
/overlaps | POST | Collisions |
/overflow | POST | Clipped / collapsed / off-screen |
/contrast | POST | WCAG ratios |
/quality | POST | Health score |
/fidelity | GET | Engine capabilities |
/health | GET | Liveness |
Every surface accepts a URL where the HTML would go. Cornea fetches the page, inlines its external stylesheets and external scripts (relative URLs resolved against the page), then inspects what a browser would actually show. That is the live coding session loop: run your dev server, point cornea at it, read the layout verdict.
cornea http://localhost:3000 390
{ "url": "http://localhost:3000", "width": 390, "height": 844 } // HTTP /inspect
{ "method": "tools/call", "params": { "name": "layout.quality",
"arguments": { "url": "http://localhost:3000", "width": 390 } } } // MCP
Honesty around capture:
height emulates a fixed viewport (screenshot frame, iframe, email) and enables below the fold clipping checks. Default 0 means an unbounded scrolling page. ┌─────────────────────────────────────┐
HTML + CSS ──► │ cornea (one binary) │ │ │ │ dom.rs html5ever ──► tree │ │ css.rs + inline │ │ │ │ layout.rs deterministic layout │ │ (block/inline/flex, │ │ box model, z-index) │ │ │ │ │ model.rs Visual Geometry Model │ │ │ │ │ inspect.rs overlap / overflow / │ │ contrast / quality │ │ │ │ │ rest.rs canonical dispatch │ └───────┬─────────────┬───────────────┘ │ │ ┌─────┴────┐ ┌────┴───────────────┐ │ CLI │ │ MCP (stdio) / HTTP│ │ cornea │ │ layout.* / REST │ └──────────┘ └────────────────────┘
Read the full technical spec: CORNEA-ARCHITECTURE.md
Cornea never silently fakes precision. layout.fidelity tells an agent exactly what it can trust, and every report carries its own warnings for sources the engine saw but did not apply (external stylesheets, external scripts, media queries, unresolved colors):
{
"exact": ["box model", "block flow", "inline text estimates", "flex row/column (no wrap)", "z-index", "visibility", "absolute/fixed left/top", "inline styles", "class/id/tag selectors", "WCAG contrast (hex, rgb, hsl, alpha)"], "approximate": ["text glyph width (not shaping)", "flex-grow/flex-basis distribution", "percentage widths", "overlap semantics ignore intentional stacking"], "deferred": ["grid (parsed as block flow)", "media queries", "border-radius", "external stylesheet when not captured", "complex selectors (combinators, pseudo)"], "js": { "engine": "boa", "phase": "A", "enabled": "opt-in via --js / js:true", "dom_shim": "static HTML mirrored in first; getElementById; innerHTML parses markup", "unsupported": ["async APIs", "event dispatch", "selector engine", "React/SPA mounting (Phase B)"] } }
Cornea can execute inline ` ` that builds its DOM via a minimal shim, then run the result through the same deterministic layout engine:
cornea page.html 360 --js # inline scripts build the DOM first
{ "html": "<p>…</p>", "width": 360, "js": true } // HTTP /inspect and MCP layout.*
document.getElementById, and innerHTML parses real markup into elements. Static content survives script runs.style.* assignments are serialized back to style="…" and participate in overlap/contrast checks.setTimeout, fetch, …) rather than hung. Any such use is surfaced in the report's js_notes, so determinism stays a guarantee. runs only when a capture layer (URL capture) inlined its body first.ROADMAP-JS.md.54 tests run clean with cargo test; CI enforces fmt, clippy `-D warnings`, release build, tests, and a CLI smoke test on every push.
$ cargo test
Running unittests src/lib.rs ... 36 passed // determinism, overlap, overflow, Running unittests src/main.rs ... 9 passed // contrast, inline flow, flex, Running tests/endpoints.rs ... 9 passed // nesting, empty input, MCP, CLI, JS, capture
layout.* is called over real stdio MCP.display:none, long text, flex row/col, inline wrapping.| Feature | Status |
|---|---|
| Block flow, box model (content/border-box) | exact |
| Inline text runs (horizontal, wrapping) | exact |
| Flex row / column (simplified, no wrap) | approximate |
| Absolute / fixed positioning (left/top) | exact |
z-index, visibility, display:none | exact |
.class / #id / tag selectors + inline styles | exact |
| WCAG contrast (hex, rgb, hsl, al
앞부분만 보여 드려요. 나머지는 저장소에서 읽어 보세요