Principal Architect Service
A comprehensive MCP server for managing product initiatives, architecture decisions, requirements, and releases, enabling AI tools to interact with architectural documentation and planning.
README
Principal Architect Service
A comprehensive Model Context Protocol (MCP) server for managing product initiatives, technical requirements, architecture decisions, and related documentation through AI development tools.
🎯 Overview
This unified MCP server provides 42+ tools across 8 domains to help software architects and product teams:
- Architecture Decision Records (ADRs) - Document and track architectural decisions with options, outcomes, and Confluence sync
- Technical Requirements - Create and manage feature requirements with Confluence publishing
- Product Requirements - Define product specifications and requirements
- Business Rule Documents - Document business rules and logic
- Technical Estimates - Track effort estimates and capacity planning
- Initiative Resources - Manage product initiatives and features
- Initiative Releases - Plan and track release schedules
- Technology Selections - Document technology choices and standards
🚀 Quick Start
Prerequisites
- Node.js 18+ (required)
- NX Monorepo tools installed
- API Server Running at
http://localhost:3000/api
Build the MCP Server
# Clone and enter the repository
cd principal-architect-service
# Install dependencies
npm install
# Build the MCP server
npx nx build product-initiatives-mcp
Start the API Server
The MCP server requires the API backend to be running:
# Start the API server (default: http://localhost:3000)
npx nx serve product-initiatives-api
📱 Installation by Tool
🖥️ Claude Desktop
Claude Desktop runs the MCP server as a subprocess and communicates via stdio.
Configuration Steps
-
Locate your Claude Desktop config file:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json - Linux:
~/.config/Claude/claude_desktop_config.json
- macOS:
-
Add the MCP server configuration:
{
"mcpServers": {
"principal-architect": {
"command": "node",
"args": [
"/absolute/path/to/principal-architect-service/dist/product-initiatives-mcp/main.js"
],
"env": {
"API_URL": "http://localhost:3000/api",
"LOG_LEVEL": "info"
}
}
}
}
Important: Replace /absolute/path/to/ with your actual project path.
- Alternative: Use npx (if in project directory)
{
"mcpServers": {
"principal-architect": {
"command": "npx",
"args": ["nx", "serve", "product-initiatives-mcp"],
"cwd": "/absolute/path/to/principal-architect-service",
"env": {
"API_URL": "http://localhost:3000/api",
"LOG_LEVEL": "info"
}
}
}
}
-
Restart Claude Desktop completely (quit and relaunch)
-
Verify Installation:
- Open Claude Desktop
- Look for the 🔌 icon in the chat interface
- You should see "principal-architect" listed with ~42 tools
- Try asking: "List all architecture decision records"
Troubleshooting Claude Desktop
Server won't start:
# Check the build output exists
ls -la dist/product-initiatives-mcp/main.js
# Test the server manually
node dist/product-initiatives-mcp/main.js
Tools not appearing:
- Open Claude Desktop Developer Tools (Help → Developer Tools)
- Check Console tab for MCP errors
- Verify API server is running:
curl http://localhost:3000/api/health
💻 Claude Code (CLI)
Claude Code can use MCP servers via stdio transport. Configuration is stored in ~/.claude/config.yaml.
Configuration Steps
- Create or edit your Claude Code config:
# Create config directory if needed
mkdir -p ~/.claude
# Edit the config file
nano ~/.claude/config.yaml
- Add MCP server configuration:
mcpServers:
principal-architect:
command: node
args:
- /absolute/path/to/principal-architect-service/dist/product-initiatives-mcp/main.js
env:
API_URL: http://localhost:3000/api
LOG_LEVEL: info
Important: Replace /absolute/path/to/ with your actual project path.
- Alternative: Run from project directory
mcpServers:
principal-architect:
command: npx
args:
- nx
- serve
- product-initiatives-mcp
cwd: /absolute/path/to/principal-architect-service
env:
API_URL: http://localhost:3000/api
LOG_LEVEL: info
- Restart Claude Code:
# If running in a session, exit and restart
# Or start a new session
claude-code
- Verify Installation:
# In Claude Code, ask:
"What MCP servers are available?"
# Test a tool:
"List all ADRs"
Using in Claude Code Session
Once configured, you can use natural language to interact with tools:
You: "Create a new ADR for choosing between REST and GraphQL APIs"
Claude: [Uses create_architecture_decision_records tool]
You: "Show me all ADRs related to authentication"
Claude: [Uses search_architecture_decision_records tool]
You: "Update ADR-001 to mark the REST option as selected"
Claude: [Uses update_architecture_decision_records tool]
Troubleshooting Claude Code
Check MCP server status:
# View available MCP servers
cat ~/.claude/config.yaml
# Test server manually
node dist/product-initiatives-mcp/main.js
Debug mode:
Set LOG_LEVEL: debug in your config to see detailed logs.
🌊 Windsurf
Windsurf supports MCP servers similar to Claude Desktop, with configuration stored in the application settings.
Configuration Steps
-
Locate Windsurf MCP config:
- macOS:
~/Library/Application Support/Windsurf/User/globalStorage/windsurf.json - Windows:
%APPDATA%\Windsurf\User\globalStorage\windsurf.json - Linux:
~/.config/Windsurf/User/globalStorage/windsurf.json
Note: Exact path may vary by Windsurf version. Check Settings → MCP Servers for the config location.
- macOS:
-
Add MCP server configuration:
{
"mcpServers": {
"principal-architect": {
"command": "node",
"args": [
"/absolute/path/to/principal-architect-service/dist/product-initiatives-mcp/main.js"
],
"env": {
"API_URL": "http://localhost:3000/api",
"LOG_LEVEL": "info"
}
}
}
}
Important: Replace /absolute/path/to/ with your actual project path.
-
Alternative: Via Windsurf UI
Some Windsurf versions support UI-based MCP configuration:
- Open Settings → Extensions → MCP Servers
- Click "Add MCP Server"
- Fill in:
- Name:
principal-architect - Command:
node - Args:
/absolute/path/to/principal-architect-service/dist/product-initiatives-mcp/main.js - Environment Variables:
API_URL=http://localhost:3000/apiLOG_LEVEL=info
- Name:
-
Reload Windsurf:
- Restart Windsurf completely
- Or use Command Palette: "Reload Window"
-
Verify Installation:
- Open the AI chat panel
- Look for MCP servers indicator
- Try: "What tools are available from the principal-architect server?"
Using in Windsurf
Windsurf integrates MCP tools into the AI coding assistant:
You: "Create an ADR for our database selection"
Windsurf: [Uses create_architecture_decision_records]
You: "Show me the ADR we created yesterday"
Windsurf: [Uses list_architecture_decision_records with filters]
Troubleshooting Windsurf
Server not loading:
- Check Windsurf's Output panel → MCP Servers
- Verify the path to main.js is correct and absolute
- Ensure API server is running
Permission issues:
# Make sure the main.js file is executable
chmod +x dist/product-initiatives-mcp/main.js
🛠️ Available Tools
Architecture Decision Records (ADRs)
8 tools for documenting architectural decisions:
| Tool | Description |
|---|---|
create_architecture_decision_records |
Create a new ADR with title, problem statement, and options |
list_architecture_decision_records |
List all ADRs (supports pagination and filtering) |
get_architecture_decision_records |
Get detailed ADR by ID including all options and outcomes |
update_architecture_decision_records |
Update ADR metadata, problem statement, or decision outcome |
update_architecture_decision_record_option |
Update specific option details, diagrams, or code examples |
search_architecture_decision_records |
Search ADRs by keyword with relevance scoring |
delete_architecture_decision_records |
Delete an ADR (soft delete preserves history) |
publish_architecture_decision_record |
Publish ADR to Confluence with ADF formatting |
Technical Requirements
5 tools for feature requirements:
| Tool | Description |
|---|---|
create_technical_requirements |
Create technical requirements document |
list_technical_requirements |
List all technical requirements |
get_technical_requirements |
Get technical requirements by ID |
update_technical_requirements |
Update technical requirements |
publish_technical_requirements_to_confluence |
Publish to Confluence |
Product Requirements
5 tools for product specs:
| Tool | Description |
|---|---|
create_product_requirements |
Create product requirements document |
list_product_requirements |
List all product requirements |
get_product_requirements |
Get product requirements by ID |
update_product_requirements |
Update product requirements |
publish_product_requirements |
Publish to Confluence |
Business Rule Documents
4 tools for business logic:
| Tool | Description |
|---|---|
create_business_rule_document |
Create business rule document |
list_business_rule_documents |
List all business rule documents |
get_business_rule_document |
Get business rule by ID |
update_business_rule_document |
Update business rule |
Technical Estimates
4 tools for effort estimation:
| Tool | Description |
|---|---|
create_technical_estimates |
Create technical estimation document |
list_technical_estimates |
List all technical estimates |
get_technical_estimates |
Get technical estimate by ID |
update_technical_estimates |
Update technical estimates |
Initiative Resources
4 tools for managing initiatives:
| Tool | Description |
|---|---|
create_product_initiative |
Create product initiative |
list_product_initiatives |
List all initiatives |
get_product_initiative |
Get initiative by ID |
update_product_initiative |
Update initiative |
Initiative Releases
5 tools for release planning:
| Tool | Description |
|---|---|
create_initiative_releases |
Create initiative release |
list_initiative_releases |
List all releases |
get_initiative_releases |
Get release by ID |
update_initiative_releases |
Update release |
publish_initiative_releases |
Publish release to Confluence |
Technology Selections
8 tools for technology decisions:
| Tool | Description |
|---|---|
create_technology_selection |
Document a technology selection |
list_technology_selections |
List all technology selections |
get_technology_selection |
Get technology selection by ID |
update_technology_selection |
Update technology selection |
delete_technology_selection |
Delete technology selection |
search_technology_selections |
Search technology selections |
filter_technology_selections_by_category |
Filter by category |
deprecate_technology_selection |
Mark technology as deprecated |
💡 Usage Examples
Example 1: Creating an ADR (Any Tool)
You: Create an ADR for choosing between MongoDB and PostgreSQL for our user database
AI: I'll create an Architecture Decision Record for the database selection...
[Uses create_architecture_decision_records tool]
Result: Created ADR-001 with:
- Title: "Database Selection for User Management"
- Problem: Need to choose a database for user data storage
- Options: MongoDB (document), PostgreSQL (relational)
- Status: Proposed
Example 2: Searching ADRs (Claude Desktop)
You: Find all ADRs related to authentication
Claude: [Uses search_architecture_decision_records tool with keyword="authentication"]
Found 3 ADRs:
1. ADR-005: OAuth vs JWT Authentication
2. ADR-012: Multi-Factor Authentication Implementation
3. ADR-018: Session Management Strategy
Example 3: Publishing to Confluence (Claude Code)
claude-code
You: Publish ADR-001 to Confluence
Claude: I'll publish the ADR to Confluence with proper formatting...
[Uses publish_architecture_decision_record tool]
✅ Published to Confluence:
- Page ID: 123456
- URL: https://your-domain.atlassian.net/wiki/spaces/ARCH/pages/123456
- Status: Published
- Stakeholders notified via @mentions
Example 4: Updating Options (Windsurf)
You: Update the MongoDB option in ADR-001 to include performance benchmarks
Windsurf: [Uses update_architecture_decision_record_option tool]
Updated option "MongoDB" with:
- Performance data: 10K writes/sec
- Latency: ~5ms p95
- Scalability: Horizontal sharding
Example 5: Creating Technical Requirements (Claude Desktop)
You: Create technical requirements for a user authentication feature
Claude: [Uses create_technical_requirements tool]
Created: TECH-REQ-001
- Feature: User Authentication
- Functional Requirements: Login, logout, password reset
- Non-functional: <2s response time, 99.9% uptime
- Dependencies: OAuth library, email service
🎨 Advanced Features
Confluence Integration
Several tools support publishing to Confluence with:
- ✅ Automatic ADF (Atlassian Document Format) conversion
- ✅ @mention resolution for stakeholders
- ✅ Linked diagrams and code examples
- ✅ Status badges and metadata
Requirements for Confluence sync:
- Confluence API credentials in API server config
- Workspace configured with Space Key
- User permissions to create/edit pages
Search and Discovery
Keyword search with relevance scoring:
- Full-text search across all ADR fields
- Highlighted snippets in results
- Relevance scores for ranking
Technology Standardization
Track your approved technology stack:
- Categorize by layer (frontend, backend, database, etc.)
- Mark technologies as approved/deprecated
- Document rationale and alternatives
- Search and filter by category
⚙️ Configuration
Environment Variables
| Variable | Default | Description |
|---|---|---|
API_URL |
http://localhost:3000/api |
Backend API base URL |
LOG_LEVEL |
info |
Logging level: error, warn, info, debug |
Custom API URL
If your API runs on a different port or host:
{
"mcpServers": {
"principal-architect": {
"env": {
"API_URL": "http://localhost:8080/api"
}
}
}
}
Debug Mode
Enable detailed logging for troubleshooting:
{
"mcpServers": {
"principal-architect": {
"env": {
"LOG_LEVEL": "debug"
}
}
}
}
🔧 Development
Project Structure
principal-architect-service/
├── apps/
│ └── product-initiatives-api/ # NestJS API backend
├── product-initiatives-mcp/ # MCP server
│ ├── src/
│ │ ├── main.ts # Server entry point
│ │ ├── tools/ # Domain-specific tools
│ │ │ ├── adrs.tools.ts
│ │ │ ├── technical-requirements.tools.ts
│ │ │ ├── product-requirements.tools.ts
│ │ │ └── ...
│ │ └── utils/ # Shared utilities
│ │ ├── logger.ts
│ │ ├── api-client.ts
│ │ └── error-handler.ts
│ └── README.md
├── libs/ # Shared libraries
│ ├── resource-schemas/ # JSON schemas
│ ├── mcp-tool-generator/ # Tool generation
│ └── confluence-client/ # Confluence integration
└── dist/ # Build output
Running Tests
# Run all tests
npm test
# Run specific domain tests
npx nx test product-initiatives-api
# Run with coverage
npm run test:coverage
# Run e2e tests
npm run test:e2e
Building for Production
# Build MCP server
npx nx build product-initiatives-mcp
# Build API server
npx nx build product-initiatives-api
# Build all
npm run build
Local Development
# Start API server (watch mode)
npx nx serve product-initiatives-api
# In another terminal, test MCP server
node dist/product-initiatives-mcp/main.js
# Or use the development build
npx nx serve product-initiatives-mcp
🐛 Troubleshooting
Common Issues
1. "Connection refused" errors
Problem: MCP server can't reach the API backend
Solution:
# Check if API is running
curl http://localhost:3000/api/health
# Start the API if needed
npx nx serve product-initiatives-api
2. "Unknown tool" errors
Problem: Tool not found or not loaded
Solution:
# Rebuild the MCP server
npx nx build product-initiatives-mcp
# Restart Claude Desktop/Code/Windsurf
3. Schema validation errors
Problem: Invalid input format
Solution:
- Check the tool's input schema in the tool list
- Ensure required fields are provided
- Verify data types match (string, number, boolean)
4. Path issues on Windows
Problem: Windows path format not recognized
Solution: Use forward slashes or escaped backslashes:
{
"command": "node",
"args": [
"C:/Users/YourName/principal-architect-service/dist/product-initiatives-mcp/main.js"
]
}
5. Port conflicts
Problem: API server port 3000 already in use
Solution:
# Use different port
PORT=3001 npx nx serve product-initiatives-api
# Update MCP config
"API_URL": "http://localhost:3001/api"
Debug Checklist
- [ ] API server is running (
curl http://localhost:3000/api/health) - [ ] MCP server is built (
ls dist/product-initiatives-mcp/main.js) - [ ] Config file has correct absolute paths
- [ ] Environment variables are set correctly
- [ ] Claude Desktop/Code/Windsurf has been restarted
- [ ] No port conflicts (check
lsof -i :3000) - [ ] Logs show successful tool initialization
Viewing Logs
Claude Desktop:
- Help → Developer Tools → Console
Claude Code:
# Set debug level in config
LOG_LEVEL: debug
# Logs output to stderr
Windsurf:
- View → Output → MCP Servers
📚 Related Documentation
🤝 Contributing
Contributions are welcome! Please follow these guidelines:
-
Run quality checks before committing:
npm run lint npm test npm run typecheck -
Follow existing patterns:
- Use NX generators for new tools
- Follow TypeScript strict mode
- Add tests for new functionality
-
Commit message format:
feat: add new ADR search capabilities fix: resolve Confluence publish error docs: update installation instructions
📄 License
MIT License
🆘 Support
Issues?
- Check the Troubleshooting section
- Review logs in your AI tool's developer console
- Verify API server is running and accessible
- Ensure all dependencies are installed
Questions?
- Review the Usage Examples
- Check tool descriptions in the Available Tools section
- Consult the MCP specification
Need Help? Open an issue on GitHub with:
- Your environment (Claude Desktop/Code/Windsurf version)
- Config file contents (sanitize sensitive data)
- Relevant logs or error messages
- Steps to reproduce the issue
🎉 Getting Started Checklist
- [ ] Clone the repository
- [ ] Run
npm install - [ ] Build the MCP server:
npx nx build product-initiatives-mcp - [ ] Start the API server:
npx nx serve product-initiatives-api - [ ] Configure your AI tool (Claude Desktop/Code/Windsurf)
- [ ] Restart your AI tool
- [ ] Test: "List all architecture decision records"
- [ ] Success! 🎊
Ready to document your architecture decisions? Start with:
"Create an ADR for our first major technical decision"
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。