什么是 ui-ticket-mcp?
连接人类与 AI 的代码评审桥梁:在浏览器审查 UI,并通过 MCP 让 agent 修复代码。
README
ui-ticket-mcp
Human-to-AI code review bridge. Review UI prototypes directly in the browser, then let AI agents read your feedback and fix the code automatically.
You click on elements, write review comments like "This button should be blue" or "The spacing is wrong here", and your AI coding agent (Claude Code, Codex, Cursor, etc.) picks them up via MCP and resolves them - with full context about which element you pointed at, its CSS, position, and surrounding DOM.
How it works
journey
title Using ui-ticket-mcp
section Review
Open your app in browser: 5: You
Click on a broken element: 4: You
Write what's wrong: 5: You
section AI resolves
Agent reads your feedback: 3: AI
Agent finds the source file: 4: AI
Agent fixes the code: 5: AI
Review disappears: 5: You, AI
One Python process handles everything - MCP protocol for the agent (stdio) and REST API for the browser UI (HTTP). Reviews are stored in a SQLite database inside your project.
Quick Start
1. Connect to your AI agent
Add to your project's .mcp.json (Claude Code, Codex, Cursor, etc.):
{
"mcpServers": {
"ui-ticket-mcp": {
"command": "uvx",
"args": ["ui-ticket-mcp"],
"env": {
"PROJECT_ROOT": "/path/to/your/project",
"REVIEW_PORT": "3200"
}
}
}
}
Restart the agent. uvx downloads and runs the package automatically - no manual install needed.
Alternative:
pip install ui-ticket-mcp, then use"command": "ui-ticket-mcp"instead ofuvx.
When the MCP server starts, it also launches a REST API on http://localhost:3200 (or your custom REVIEW_PORT) for the browser UI.
Important: The API always runs locally (
localhost). The websitehttps://uiticket.0ics.ai/is the landing page only — it is NOT an API endpoint. Always usehttp://localhost:{PORT}/apias theapi-url.
2. Add the browser UI to your app
npm install ui-ticket-panel
In your app's entry file (e.g. main.ts, index.tsx):
import { defineReviewPanel } from 'ui-ticket-panel';
defineReviewPanel();
Then in your root template:
<review-panel api-url="http://localhost:3200/api"></review-panel>
That's it. Works in any framework - Angular, React, Vue, Svelte, or plain HTML. It's a standard Web Component. For SSR frameworks (Next.js, Nuxt, SvelteKit) see the Framework Examples section — you need a dynamic import on the client side.
No bundler? Use CDN
<script type="module" src="https://unpkg.com/ui-ticket-panel/dist/bundle.js"></script>
<review-panel api-url="http://localhost:3200/api"></review-panel>
The bundle auto-registers the <review-panel> element. No npm install, no build step needed.
3. Start reviewing
Open your app in the browser. You'll see a floating chat button in the bottom-right corner. Click it to open the review panel, or press Alt+A to enter annotation mode and click directly on elements.
Browser UI Features
Review Panel
The floating panel lets you browse, filter, and manage all reviews:
- Filter tabs - Switch between Open, Resolved, and All reviews
- Search - Full-text search across all review comments
- Tag filter - Filter by category: general, bug, suggestion, question
- Per-review actions - Resolve, Reopen, Delete, Reply, Highlight element
- Threaded replies - Reply to reviews for back-and-forth discussion
- Manual review form - Write reviews without annotation (Ctrl+Enter to submit)
- Badge counter - Floating button shows count of open reviews
Annotation System
The annotation system lets you point at specific elements and attach reviews to them:
- Click-to-annotate - Press Alt+A (or the target button), then click any element
- Multi-select drag - Click and drag to select a region of multiple elements
- Hover preview - See element identification in real-time as you move the mouse
- Smart popup - Appears above or below the element depending on available space
When you annotate an element, the system captures rich metadata that helps the AI agent understand exactly what you're pointing at:
| Captured data | Example |
|---|---|
| Element name | Button 'Save', Input[email] 'Your email', Heading 2 'Features' |
| CSS selector | #main-header, div.card > button.primary:nth-of-type(2) |
| Bounding box | Position and dimensions in pixels |
| Nearby text | Own text + previous/next sibling text for context |
| Selected text | If you highlight text before annotating |
| CSS classes | Filtered (excludes framework-generated hashes) |
| Computed styles | Color, background, font, border, padding (smart per element type) |
| Full DOM path | body > div#app > section.content > div.card > button |
| Accessibility | ARIA roles, labels, tabindex, focusability |
| Sibling context | Parent tag, children count, adjacent sibling tags |
Marker Badges
Reviews with annotations show numbered badges on the page next to the annotated element:
- Single review - Circular badge with the review ID, colored red (open) or green (resolved)
- Stacked reviews (3+ on the same element) - Pill badge showing count, with a gradient showing open/resolved ratio
- Click badge - Opens the review in the panel
- Delete badge - Remove via the X button on hover
- Tooltip - Hover to see author, element name, and comment preview
Tags
Every review can be tagged with a category:
| Tag | Color | Use for |
|---|---|---|
general | Indigo | General feedback (default) |
bug | Red | Something is broken |
suggestion | Green | Improvement idea |
question | Amber | Needs clarification |
Keyboard Shortcuts
| Shortcut | Action |
|---|---|
| Alt+A | Toggle annotation mode |
| Ctrl+Enter | Submit review or reply |
| Escape | Close popup / exit annotation mode |
Architecture
graph LR
Agent[AI Agent]
Browser[Reviewer - Browser]
Server[ui-ticket-mcp]
DB[(SQLite)]
Agent <-->|stdio MCP| Server
Browser <-->|HTTP REST :3200| Server
Server --- DB
- MCP (stdio) - Your agent framework starts it automatically. 10 tools for AI agents to read, resolve, and manage reviews.
- REST API (HTTP :3200) - Starts in background, serves the browser review UI. CORS enabled for all origins.
- SQLite (WAL mode) - Concurrent readers + 1 writer, 5s busy timeout. Database lives inside your project at
.reviews/reviews.db.
MCP Tools
10 tools available to AI agents:
get_review_summary()
Overview of all pages with review counts.
Page | Open | Resolved | Total
------------ | ---- | -------- | -----
user-profile | 3 | 1 | 4
dashboard | 0 | 2 | 2
get_reviews(page_id?: str)
List review comments. Optionally filtered by page. Shows status, tag, element context, and reply chains.
[#1] [OPEN] [bug] user-profile - alice: The header spacing is off
→ Element: Heading 2 'User Profile' | Selector: h2.page-title
[#2] [RESOLVED] user-profile - bob: Button color should be blue
get_annotated_reviews(page_id?: str)
Returns only reviews that have element annotation metadata. Includes element name, CSS selector, full DOM path, selected text, accessibility info - everything the agent needs to locate and understand the annotated element.
get_pending_work()
All open reviews grouped by page - the agent's "todo list".
## user-profile (2 open)
- #1 [bug] (alice): The header spacing is off
- #3 [suggestion] (alice): Add hover state to buttons
## dashboard (1 open)
- #4 (bob): Chart labels are truncated
add_review(page_id, author, text, tag?, metadata?, parent_id?)
Create a new review. Supports tags, annotation metadata (JSON), and threading via parent_id.
resolve_review(review_id, resolved_by?)
Mark a review as resolved. Sets resolved_at timestamp and resolved_by (defaults to "agent").
reopen_review(review_id)
Reopen a previously resolved review. Clears resolution info.
batch_resolve(page_id, resolved_by?)
Resolve all open reviews on a page at once. Returns Resolved 3 review(s) on user-profile.
find_source_file_tool(page_id)
Find source files in PROJECT_ROOT matching a page ID. Searches by kebab-case, CamelCase, and glob patterns. Skips node_modules, dist, .git.
Found 3 file(s) for 'user-profile':
- src/app/user-profile/user-profile.component.ts
- src/app/user-profile/user-profile.component.html
- src/app/shared/UserProfile.ts
get_setup_guide()
Returns the full setup guide (MCP config, REST API, browser UI). Useful when the agent needs to help set up the review system in a new project.
Typical agent workflow
graph TD
A["get_pending_work()"] -->|See what needs attention| B["get_annotated_reviews(page)"]
B -->|Get element metadata for context| C["find_source_file_tool(page)"]
C -->|Locate the source files| D["Read & edit the code"]
D --> E{Resolve}
E -->|Single| F["resolve_review(id)"]
E -->|All on page| G["batch_resolve(page)"]
REST API
All endpoints under /api. CORS enabled for all origins.
Reviews
| Method | Endpoint | Description |
|---|---|---|
| GET | /api/reviews/summary | Per-page summary with open/resolved counts |
| GET | /api/reviews | All reviews (newest first) |
| GET | /api/reviews/{page_id} | Reviews for a page. Query: ?status=open|resolved, ?tag=bug|suggestion|... |
| POST | /api/reviews/{page_id} | Create review |
| PATCH | /api/review/{id} | Update review (status, text, tag, metadata) |
| DELETE | /api/review/{id} | Delete review permanently |
Replies
| Method | Endpoint | Description |
|---|---|---|
| GET | /api/review/{id}/replies | Get all replies to a review (chronological) |
| POST | /api/reviews/{page_id} | Create reply (include parent_id in body) |
POST body
| Field | Type | Default | Description |
|---|---|---|---|
text | string | - | Review comment (required) |
author | string | "anonymous" | Reviewer name |
tag | string | "general" | "general", "bug", "suggestion", or "question" |
metadata | object | - | Annotation context (element, selector, styles, etc.) |
parent_id | integer | - | Parent review ID for threaded replies |
PATCH body
| Field | Type | Description |
|---|---|---|
status | "open" | "resolved" | Resolving auto-sets resolved_at and resolved_by |
text | string | Updated comment text |
tag | string | Updated tag |
resolved_by | string | Who resolved it (default: "user" via API, "agent" via MCP) |
metadata | object | Updated annotation metadata |
Database
Reviews are stored in SQLite inside your project at {PROJECT_ROOT}/.reviews/reviews.db. The database is auto-created on first run.
The .reviews/ directory includes:
| File | Purpose |
|---|---|
reviews.db | SQLite database (commit to git to share reviews with your team) |
.gitkeep | Ensures directory is tracked |
.gitignore | Ignores WAL temp files (*.db-wal, *.db-shm) |
Path resolution:
REVIEW_DB_PATHenv var (explicit override)PROJECT_ROOT/.reviews/reviews.db(default)./reviews.db(fallback)
Schema
reviews (
id INTEGER PRIMARY KEY,
page_id TEXT NOT NULL,
author TEXT DEFAULT 'anonymous',
text TEXT NOT NULL,
status TEXT DEFAULT 'open', -- 'open' | 'resolved'
created_at TEXT NOT NULL, -- ISO 8601
resolved_at TEXT,
resolved_by TEXT,
metadata TEXT, -- JSON: annotation context
tag TEXT DEFAULT 'general', -- 'general' | 'bug' | 'suggestion' | 'question'
parent_id INTEGER REFERENCES reviews(id) -- threaded replies
)
Annotation metadata (JSON)
When a review is created via annotation, the metadata field contains:
{
"element": "Button 'Save'",
"selector": "button.btn-primary",
"boundingBox": { "x": 100, "y": 200, "width": 80, "height": 40 },
"selectedText": "Click to save",
"cssClasses": "btn btn-primary active",
"nearbyText": "Save your work | [after:] Cancel",
"nearbyElements": "Parent: form.editor (5 children) | Siblings: input, button.secondary",
"computedStyles": "color: #fff, background: #3b82f6, border-radius: 4px",
"fullPath": "body > div#app > div.modal > form > button",
"accessibility": "role=\"button\", tabindex=\"0\", focusable",
"isMultiSelect": false,
"url": "http://localhost:4200/user-profile"
}
This metadata gives the AI agent precise context about what you annotated - which element, where it is, what it looks like, and how to find it in the DOM.
Web Component Attributes
| Attribute | Required | Description |
|---|---|---|
api-url | Yes | REST API base URL (e.g. http://localhost:3200/api) |
page-id | No | Explicit page identifier for filtering reviews. If omitted, auto-detection is used (recommended) |
Page identification
The panel needs to know which page the user is on, so it can show and file reviews for that specific page. There are two modes:
Automatic detection (recommended)
When no page-id attribute is set, the panel derives the page identifier from the URL pathname:
| URL | Page ID |
|---|---|
/ | home |
/analytics | analytics |
/settings | settings |
/user/profile | user/profile |
The panel also listens for SPA navigation events (pushState, replaceState, popstate) and automatically reloads reviews when the route changes. This means it works out of the box with client-side routing in React Router, Vue Router, Angular Router, Next.js, etc.
<!-- Auto-detection: no page-id attribute needed -->
<review-panel api-url="http://localhost:3200/api"></review-panel>
Explicit page ID
If you need to control the page ID yourself (e.g. your pages don't map cleanly to URL paths), set the page-id attribute:
<review-panel api-url="http://localhost:3200/api" page-id="dashboard"></review-panel>
Important: These two modes are mutually exclusive. When
page-idis set, auto-detection is completely disabled — the panel will NOT react to route changes. Do not combine both.
Programmatic API
const panel = document.querySelector('review-panel');
// Change page without reloading
panel.setPageId('dashboard');
Packages
| Package | Registry | Description |
|---|---|---|
ui-ticket-mcp | PyPI | Python MCP server + REST API |
ui-ticket-panel | npm | <review-panel> Web Component |
ui-ticket-core | npm | Framework-agnostic core: types, API client, reactive store, annotation engine |
Environment Variables
| Variable | Default | Description |
|---|---|---|
PROJECT_ROOT | - | Root of the reviewed project. DB auto-created at {PROJECT_ROOT}/.reviews/ |
REVIEW_DB_PATH | (auto) | Explicit DB path override. Takes priority over PROJECT_ROOT |
REVIEW_PORT | 3200 | Port for the REST API server |
Framework Examples
With a bundler
In your entry file (e.g. main.ts, main.js):
import { defineReviewPanel } from 'ui-ticket-panel';
defineReviewPanel();
Then in your HTML:
<review-panel api-url="http://localhost:3200/api"></review-panel>
Plain HTML (no bundler / CDN)
<script type="module" src="https://unpkg.com/ui-ticket-panel/dist/bundle.js"></script>
<review-panel api-url="http://localhost:3200/api"></review-panel>
React
import { defineReviewPanel } from 'ui-ticket-panel';
defineReviewPanel();
function App() {
return <review-panel api-url="http://localhost:3200/api" />;
}
Vue
<template>
<review-panel api-url="http://localhost:3200/api"></review-panel>
</template>
<script setup>
import { defineReviewPanel } from 'ui-ticket-panel';
defineReviewPanel();
</script>
Angular
// app.config.ts
import { defineReviewPanel } from 'ui-ticket-panel';
defineReviewPanel();
// component - add CUSTOM_ELEMENTS_SCHEMA
@Component({
schemas: [CUSTOM_ELEMENTS_SCHEMA],
template: `<review-panel api-url="http://localhost:3200/api"></review-panel>`
})
Svelte
<script>
import { defineReviewPanel } from 'ui-ticket-panel';
defineReviewPanel();
</script>
<review-panel api-url="http://localhost:3200/api"></review-panel>
Next.js (SSR)
Web Components use window and HTMLElement which don't exist during server-side rendering. You must load the panel dynamically on the client side:
// components/ReviewPanel.tsx
'use client';
import { useEffect } from 'react';
export default function ReviewPanel() {
useEffect(() => {
import('ui-ticket-panel').then(m => m.defineReviewPanel());
}, []);
return <review-panel api-url="http://localhost:3200/api" />;
}
Then use it in your root layout:
// app/layout.tsx
import ReviewPanel from './components/ReviewPanel';
export default function RootLayout({ children }) {
return (
<html>
<body>
{children}
<ReviewPanel />
</body>
</html>
);
}
Nuxt (SSR)
<template>
<ClientOnly>
<review-panel api-url="http://localhost:3200/api"></review-panel>
</ClientOnly>
</template>
<script setup>
import { onMounted } from 'vue';
onMounted(async () => {
const { defineReviewPanel } = await import('ui-ticket-panel');
defineReviewPanel();
});
</script>
SvelteKit (SSR)
<script>
import { onMount } from 'svelte';
onMount(async () => {
const { defineReviewPanel } = await import('ui-ticket-panel');
defineReviewPanel();
});
</script>
<review-panel api-url="http://localhost:3200/api"></review-panel>
Issues & Feedback
Found a bug or have a feature request? Open an issue on this repository.
License
CC BY-NC 4.0 — free for study, research, and non-commercial use. See LICENSE for details.
Built by Šimon Cmar, Ladislav Sopko & Lorenzo Leoni
常见问题
ui-ticket-mcp 是什么?
连接人类与 AI 的代码评审桥梁:在浏览器审查 UI,并通过 MCP 让 agent 修复代码。
相关 Skills
MCP构建
by anthropics
聚焦高质量 MCP Server 开发,覆盖协议研究、工具设计、错误处理与传输选型,适合用 FastMCP 或 MCP SDK 对接外部 API、封装服务能力。
✎ 想让 LLM 稳定调用外部 API,就用 MCP构建:从 Python 到 Node 都有成熟指引,帮你更快做出高质量 MCP 服务器。
Slack动图
by anthropics
面向Slack的动图制作Skill,内置emoji/消息GIF的尺寸、帧率和色彩约束、校验与优化流程,适合把创意或上传图片快速做成可直接发送的Slack动画。
✎ 帮你快速做出适配 Slack 的动图,内置约束规则和校验工具,少踩上传与播放坑,做表情包和演示都更省心。
MCP服务构建器
by alirezarezvani
从 OpenAPI 一键生成 Python/TypeScript MCP server 脚手架,并校验 tool schema、命名规范与版本兼容性,适合把现有 REST API 快速发布成可生产演进的 MCP 服务。
✎ 帮你快速搭建 MCP 服务与后端 API,脚手架完善、扩展顺手,尤其适合想高效验证服务能力的开发者。
相关 MCP Server
Slack 消息
编辑精选by Anthropic
Slack 是让 AI 助手直接读写你的 Slack 频道和消息的 MCP 服务器。
✎ 这个服务器解决了团队协作中需要 AI 实时获取 Slack 信息的痛点,特别适合开发团队让 Claude 帮忙汇总频道讨论或发送通知。不过,它目前只是参考实现,文档有限,不建议在生产环境直接使用——更适合开发者学习 MCP 如何集成第三方服务。
by netdata
io.github.netdata/mcp-server 是让 AI 助手实时监控服务器指标和日志的 MCP 服务器。
✎ 这个工具解决了运维人员需要手动检查系统状态的痛点,最适合 DevOps 团队让 Claude 自动分析性能数据。不过,它依赖 NetData 的现有部署,如果你没用过这个监控平台,得先花时间配置。
by d4vinci
Scrapling MCP Server 是专为现代网页设计的智能爬虫工具,支持绕过 Cloudflare 等反爬机制。
✎ 这个工具解决了爬取动态网页和反爬网站时的头疼问题,特别适合需要批量采集电商价格或新闻数据的开发者。不过,它依赖外部浏览器引擎,资源消耗较大,不适合轻量级任务。