why-ui

Coding agents can read CSS. why-ui lets them inspect what the browser actually did.

Find the element covering a button, trace it toward local source, and independently verify the fix after the agent edits code.

Actual Chromium + production extension + MCP. Scripted fixture source edit. 100% sampled blockage → 0%. Evidence and reproduction.

CI · Release · Framework checks · Security

Quick start

Requires Node 22.13+ and desktop Chrome 116+ with unpacked extensions allowed. CI uses Chromium 153, Linux/Node 22, and Windows/Node 24.

1. Download the extension ZIP. Extract it into a stable directory. In chrome://extensions, enable Developer mode, select Load unpacked, and choose that directory. Pin the why-ui action.

2. Install the GitHub release and pair:

   npm install -g https://github.com/Yudis-bit/why-ui/releases/download/v0.1.0/why-ui-0.1.0.tgz
   why-ui pair

   Open extension Options and enter the displayed one-time token and port.

3. Configure your agent's MCP connection below. Open your application in Chrome, select the why-ui action or press Alt+Shift+Y, then move the pointer over the interaction.

why-ui doctor --workspace /path/to/app checks the local installation and prints an absolute-path MCP configuration, including for Windows. It never prints credentials. TCP reachability alone does not prove a connected extension or armed tab.

Downloads include SHA256SUMS.txt. Distribution is GitHub-first; no npm-registry publication is claimed. Prefer source builds? Instructions below.

Related MCP server: uisight-mcp

Claude Code / Cursor

For Claude Code, run from the application repository:

claude mcp add --transport stdio --scope local why-ui -- why-ui mcp --workspace /path/to/app

Use /mcp to check the connection. If your agent cannot resolve the global command (particularly a Windows npm shim), use the absolute Node command printed by why-ui doctor.

For Cursor, put this in the application's .cursor/mcp.json. Replace the paths with those printed by why-ui doctor --workspace /path/to/app --json under mcpServer:

{
  "mcpServers": {
    "why-ui": {
      "command": "node",
      "args": ["/path/to/why-ui/bin/why-ui.js", "mcp", "--workspace", "/path/to/app"]
    }
  }
}

Claude Code configuration · Cursor configuration. Client formats are documented; automated transport tests use the official MCP SDK. One daemon, one agent connection, and one armed tab are supported at a time.

Inspect → patch → verify

Call inspect_interaction with {} after pointing at a reachable part of the target. A fully covered or pointer-transparent target needs a known identity:

{ "target": { "selector": "#payment" }, "maxSamples": 64 }

The tool returns a concise human-readable summary and structured evidence: target, primary blocker, sampled surface, diagnosis, stacking explanation, local source references with explicit certainty, limitations, and an inspectionId.

After the agent edits source and the application updates:

{ "inspectionId": "<returned inspectionId>", "stabilizationTimeoutMs": 10000 }

Call this with verify_fix:

• VERIFIED_PASS: identity reconciles, source/runtime changed, fresh geometry and hits stabilize, the diagnosed failure is absent, and a meaningful Safe Core exists.

• VERIFIED_FAIL: fresh browser evidence still proves a failure, including a new blocker.

• VERIFY_INCONCLUSIVE: identity, readiness, or safe-region evidence is insufficient.

why-ui does not edit code or activate controls. Reflow is normal: old geometry does not establish the result. Verification semantics and thresholds.

How it works

flowchart LR
  Chrome[Armed Chrome tab] --> Sensor[MAIN-world sensor]
  Sensor --> Extension[MV3 extension]
  Extension <-->|Authenticated loopback WebSocket| Daemon[Local daemon]
  Daemon <-->|MCP stdio| Agent[Coding agent]
  Agent -->|Source patch| App[Application]
  App -->|HMR / reload| Chrome
  Daemon --> Source[Workspace source resolution]

Browser hits establish interaction truth. why-ui explains relevant stacking contexts and maps bounded runtime hints toward local files. The agent reasons about the patch; why-ui independently reruns runtime assertions. Architecture.

Security and limits

Loopback 127.0.0.1 only. Single-use pairing binds an extension Origin and a 32-byte secret; sessions use HMAC challenge-response. Auth lives in ~/.why-ui/auth.json, mode 0600 where supported. Human logs use stderr; MCP stdout is protocol only.

No cookies, page-storage values, input/textarea contents, full HTML/DOM, screenshots, network bodies, or Fiber props/state are collected during inspection. Bounded labels, IDs, classes, keys and source paths remain evidence visible to the agent. Threat model · Private vulnerability reporting.

React is optional. Tested development setups include Vite/React 18 and Next App Router / Turbopack / React 19 native HMR. Exact versions and limitations. Closed shadow interiors and iframe contents remain opaque. Missing maps become UNMAPPED; static correlations are heuristic. Source-map fetching stays on the app's same loopback origin. Ratios and Safe Core are estimates; handler correctness, keyboard behavior and exact geometric coverage are not proved.

If pairing expires, restart why-ui pair. For lost pairing, stop the MCP process, run why-ui pair --reset, choose Forget pairing in Options, and pair again. Default port is 9876; set --port and the Options port together if occupied. Re-arm after cross-origin navigation. Keep the unpacked extension path stable.

Build from source

git clone https://github.com/Yudis-bit/why-ui.git
cd why-ui
npm ci
npm run build
node bin/why-ui.js pair

Load dist/extension/ unpacked and configure the agent to run node with this checkout's absolute bin/why-ui.js path. npm link optionally installs the why-ui command.

Development

npx playwright install --with-deps chromium
npm run typecheck
npm test
npm run test:browser
npm run test:e2e
npm run test:dogfood
npm run test:package
npm run build:extension

Contributing · Browser validation · Bridge protocol · Share a UI bug. Licensed under MIT.