PRD Builder MCP
Enables product managers to create, edit, and export product requirements documents interactively within Claude's chat, with real-time completeness scoring and suggestions.
README
PRD Builder — MCP App
An interactive PRD (Product Requirements Document) builder that renders inside Claude's chat. Built on the MCP Apps extension.
What This Is
An MCP server + interactive React UI that gives product managers a structured, visual PRD builder embedded directly in their Claude conversation. The PM chats with Claude to draft content, while a real-time visual interface shows section completion status, quality scores, and actionable suggestions.
This is not a text-only tool. Unlike existing PRD skills and MCP integrations, this renders an interactive UI inline in the chat — the PM can edit sections, reorder them, see completeness feedback, and export, all without leaving the conversation.
The Experience
PM: "Help me write a PRD for the onboarding redesign"
Claude: *calls open_prd_builder → interactive UI renders inline*
┌─────────────────────────────────────────────┐
│ Untitled PRD [Template ▼] [Export] │
├─────────────┬───────────────────────────────┤
│ Sections │ Problem Statement │
│ │ ┌─────────────────────────┐ │
│ ⬜ Title* │ │ [editable textarea] │ │
│ ⬜ Problem* │ │ │ │
│ ⬜ Users* │ └─────────────────────────┘ │
│ ⬜ Solution* │ ⚠️ Only 12 chars — aim for 50+│
│ ⬜ Stories* ├───────────────────────────────┤
│ ⬜ Reqs* │ Completeness: 0% │
│ ⬜ Metrics* │ ⬜ 8 empty 🔄 0 👀 0 ✅ 0 │
│ ⬜ Timeline │ 💡 Start with: "What problem │
│ ⬜ Risks │ are users experiencing?" │
└─────────────┴───────────────────────────────┘
PM: "The problem is that 67% of users abandon signup at step 3"
Claude: *drafts problem statement, calls update_prd_section*
→ Content appears in the UI, status changes ⬜→🔄, score updates
Project Structure
prd-builder-mcp/
├── packages/
│ ├── shared/ # Core domain types (the contract)
│ │ └── src/types.ts
│ │
│ ├── prd-engine/ # Business logic (no MCP/UI dependencies)
│ │ ├── src/
│ │ │ ├── templates/ # PRD template definitions
│ │ │ │ └── default-prd.ts # 3 templates: feature, platform, bugfix
│ │ │ ├── validation/ # Section validation rules
│ │ │ │ └── section-rules.ts # Text, list, user-story, metrics rules
│ │ │ ├── scoring/ # Completeness scoring engine
│ │ │ │ └── completeness.ts # Weighted scoring + suggestions
│ │ │ ├── factory.ts # Create PRD from template, update/reorder
│ │ │ ├── export.ts # Markdown, JSON, plain text export
│ │ │ └── index.ts # Public API
│ │ └── package.json
│ │
│ ├── mcp-server/ # MCP protocol layer
│ │ ├── src/
│ │ │ ├── tools/ # Tool definitions with UI metadata
│ │ │ │ └── prd-tools.ts # 5 tools: open, analyze, update, export, list
│ │ │ ├── handlers/ # Tool call handlers
│ │ │ │ ├── prd-handlers.ts # Business logic per tool
│ │ │ │ └── store.ts # In-memory PRD store (swap for persistent)
│ │ │ ├── resources/ # UI resource serving (ui:// scheme)
│ │ │ │ └── ui-resources.ts # Returns HTML for iframe rendering
│ │ │ └── index.ts # Server entry — stdio transport
│ │ └── package.json
│ │
│ └── ui/ # Interactive React frontend
│ ├── src/
│ │ ├── hooks/ # MCP communication + state management
│ │ │ ├── useMCPApp.ts # postMessage bridge to host/server
│ │ │ └── usePRDState.ts # Optimistic local state
│ │ ├── components/ # Visual components
│ │ │ ├── Toolbar.tsx # Title, template selector, export
│ │ │ ├── PRDOutline.tsx # Sidebar: section list + drag reorder
│ │ │ ├── SectionEditor.tsx# Inline editing + validation feedback
│ │ │ └── CompletenessMeter.tsx # Score gauge + breakdown
│ │ ├── App.tsx # Main layout wiring everything together
│ │ ├── main.tsx # React entry point
│ │ └── styles/main.css # Self-contained styling
│ └── package.json
│
├── docs/
│ ├── DEPLOYMENT.md # Comprehensive guide: local dev → production
│ └── ARCHITECTURE.md # Technical architecture deep-dive
│
├── package.json # Root workspace config
├── tsconfig.json # TypeScript config with path aliases
└── vite.config.ts # Vite build config for UI
MCP Tools
| Tool | UI? | Description |
|---|---|---|
open_prd_builder |
✅ | Opens the interactive PRD builder UI inline |
analyze_prd |
— | Returns completeness score + suggestions (text only) |
update_prd_section |
— | Pushes content into a section (used by Claude) |
export_prd |
— | Exports PRD to markdown/JSON |
list_prd_templates |
— | Lists available templates |
PRD Templates
| Template | Sections | Use Case |
|---|---|---|
| Standard Feature | 11 | New product features (default) |
| Platform Initiative | 9 | Infrastructure / cross-cutting changes |
| Bugfix / Small Change | 5 | Lightweight documentation for fixes |
Getting Started
# Install
npm install
# Build
npm run build
# Configure in Claude Desktop
# See docs/DEPLOYMENT.md → Phase 3
Local deployment guide: docs/DEPLOYMENT.md
External hosting (Railway): docs/DEPLOY-RAILWAY.md
Architecture deep-dive: docs/ARCHITECTURE.md
Tech Stack
- MCP SDK:
@modelcontextprotocol/sdk+@modelcontextprotocol/ext-apps - Server: Node.js + TypeScript
- UI: React 18 + Vite
- Transport: stdio (local) / SSE (remote)
- No runtime database — in-memory store (persistent storage is a drop-in replacement, see DEPLOYMENT.md)
Status
Prototype — core code is complete and functional. Not yet built or tested against a live Claude instance. The DEPLOYMENT.md guide walks through everything needed to take it live.
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。