Nitpick
A MCP server that enables AI coding agents to consume structured UI feedback via click-to-annotate, supporting issue types and severity levels.
README
Nitpick
MCP server + Chrome extension for structured, click-to-annotate UI feedback that AI coding agents (Cursor, Claude Code, etc.) can consume as context.
Status: Week 1 — MCP server core. Chrome extension lands in weeks 3–4.
What's here
| Package | Role |
|---|---|
@nitpick/shared |
Zod feedback schema (single source of truth) |
@nitpick/mcp-server |
MCP tools over stdio + localhost HTTP ingest |
Prerequisites
- Node.js 20+
- npm 10+
Setup
git clone https://github.com/salahashraf3/Nitpick.git
cd Nitpick
npm install
npm run build
Run the MCP server
npm run start:server
This starts:
- MCP stdio — for Cursor / Claude Code
- HTTP ingest on
http://127.0.0.1:3847(default) — for the future Chrome extension
If 3847 is already taken, the server picks the next free port (3848, 3849, …) and logs the URL on stderr. To pin a port:
NITPICK_HTTP_PORT=3850 npm run start:server
If you set NITPICK_HTTP_PORT and that port is busy, startup fails with a clear error (no silent fallback), so the extension can keep a fixed URL.
Health check (use the port from the log line):
curl http://127.0.0.1:3847/health
Submit feedback via HTTP (hand-crafted payload):
curl -s -X POST http://127.0.0.1:3847/feedback \
-H 'Content-Type: application/json' \
-d '{
"url": "https://app.example.com/dashboard",
"target": {
"selector": "#dashboard > div.card-header:nth-child(2)",
"tagName": "div",
"textContent": "Revenue Overview",
"boundingBox": { "x": 120, "y": 84, "width": 320, "height": 48 }
},
"issueType": "spacing",
"severity": "minor",
"description": "Card header has too much bottom padding compared to other cards"
}'
Feedback is kept in memory and persisted to ~/.nitpick/feedback.json.
Cursor MCP config
Add to your Cursor MCP settings (~/.cursor/mcp.json or project .cursor/mcp.json):
{
"mcpServers": {
"nitpick": {
"command": "node",
"args": ["/ABSOLUTE/PATH/TO/Nitpick/packages/mcp-server/dist/index.js"]
}
}
}
Replace /ABSOLUTE/PATH/TO/Nitpick with your clone path. After reload, the agent can call submit_ui_feedback.
Claude Code
claude mcp add nitpick -- node /ABSOLUTE/PATH/TO/Nitpick/packages/mcp-server/dist/index.js
MCP tool (week 1)
submit_ui_feedback
Required: target, issueType, description.
Optional: url, severity (default minor), screenshot, suggestedFix, domContext.
issueType: spacing | alignment | color | typography | responsiveness | interaction | content | other
severity: blocker | major | minor | nit
Project layout
packages/
shared/ # Zod schema + types
mcp-server/ # MCP + HTTP ingest
chrome-extension/ # (weeks 3–4)
docs/
Roadmap
- Weeks 1–2: MCP server core ← you are here
- Weeks 3–4: Chrome extension capture
- Weeks 5–6: Extension → server end-to-end
- Week 7: Polish + demo GIF
- Weeks 8–9: Resume packaging + soft launch
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。