GitHubスター
35
ユーザー評価
未評価
お気に入り
0
閲覧数
4
フォーク
7
イシュー
1
ZMCPTools
🚀 TypeScript MCP Tools for Claude Code - Professional multi-agent orchestration platform with 61 enhanced tools, documentation intelligence, and advanced automation capabilities.
⚠️ Important Setup Note
Before spawning agents, run this command once to enable proper agent permissions:
claude --dangerously-skip-permissions
Agents run on daemon threads and need this permission to execute properly.
✨ Key Features
🎯 Multi-Agent Orchestration
- Architect-Led Coordination: AI architect automatically spawns and coordinates specialized agent teams
- Intelligent Dependencies: Agents work in proper order (Backend → Frontend → Testing → Documentation)
- Real-Time Communication: Agents collaborate through dedicated chat rooms with message broadcasting
- Foundation Session Caching: 85-90% cost reduction through automatic shared context management
- Professional Task Management: Create, assign, track, and monitor complex development workflows
🎨 TypeScript-First Architecture
- Type-Safe MCP Server: Built with Zod schemas and strict TypeScript for reliability
- Modern CLI Interface: Commander.js-powered CLI with structured command hierarchy
- Development Ready: One-command setup with hot-reload development via tsx
- Binary Distribution: Global access via
claude-mcp-tools
andclaude-mcp-server
commands - Professional Build System: tsup-based compilation with dual CLI/server binaries
🌐 Advanced Browser Automation
- Playwright Integration: Professional web automation with session management
- AI-Powered DOM Analysis: Intelligent page structure analysis and navigation
- Screenshot Analysis: AI-driven visual page analysis with region focusing
- Smart Session Management: Auto-cleanup, session persistence, and connection pooling
- Legacy Support: Comprehensive tool migration with backward compatibility
📚 Documentation Intelligence & Vector Search
- LanceDB Vector Database: Local, high-performance semantic search with multiple embedding providers
- Intelligent Web Scraping: Multi-page documentation collection with automatic vectorization
- Advanced Content Processing: Smart URL filtering, pattern matching, and content extraction
- Job Management: Background worker system with status monitoring and job control
- Documentation Sources: Track and manage multiple documentation repositories
🧠 Knowledge Graph & Memory Systems
- Graph-Based Knowledge Storage: Entity-relationship modeling for cross-agent learning
- Semantic Search: Vector-powered knowledge discovery and relationship traversal
- Shared Memory: Persistent agent collaboration and insight sharing
- Project Analysis: Comprehensive code structure analysis with symbol extraction
- Smart File Operations: Pattern-based file operations with fuzzy matching
🚀 Quick Installation
Prerequisites
- Node.js 18+: Required for TypeScript runtime and LanceDB native bindings
- Claude Code CLI: Anthropic's Claude Code CLI
- Package Manager: npm (included), yarn, pnpm, or bun
Production Installation (Recommended)
# Install globally first (recommended for WSL/Linux compatibility)
pnpm add -g zmcp-tools
# If requested, approve build scripts for native dependencies
pnpm approve-builds -g
# Then install MCP integration
zmcp-tools install
# Alternative: Direct installation (may have issues with Sharp in WSL)
npx zmcp-tools@latest install # npm
yarn dlx zmcp-tools@latest install # yarn
bunx zmcp-tools@latest install # bun
This automatically:
- ✅ Installs MCP server to
~/.mcptools/server/
- ✅ Configures Claude Code with
claude mcp add --scope local
(current directory only) - ✅ Sets up project permissions and CLAUDE.md integration
- ✅ Initializes SQLite database for agent coordination
- ✅ Initializes LanceDB vector database for semantic search
- ✅ Creates 61 professional MCP tools ready for use
Development Installation
# Clone and setup development environment
git clone https://github.com/zachhandley/ZMCPTools
cd ZMCPTools
# Quick automated setup
pnpm install && pnpm run install:global
# Or manual setup
pnpm install # Install dependencies
pnpm build # Compile TypeScript
pnpm link --global # Create global symlink
zmcp-tools install # Configure MCP integration
Development features:
- ✅ Global
zmcp-tools
command - ✅ Hot-reload development:
pnpm dev
- ✅ TypeScript compilation:
pnpm build
- ✅ Test suite:
pnpm test
- ✅ Full source code access and modification
MCP Server Configuration
The installer automatically configures the MCP server using claude mcp add --scope local
. The server runs directly with Node.js:
# Automatically executed during installation
claude mcp add --scope local zmcp-tools node ~/.mcptools/server/index.js
This provides:
- ✅ Core MCP server with 43 tools (including LanceDB)
- ✅ Multi-agent orchestration capabilities
- ✅ TypeScript type safety and performance
- ✅ SQLite-based data persistence
- ✅ LanceDB vector database for semantic search
- ✅ Advanced file operations and project analysis
- ✅ Documentation intelligence with vector embeddings
- ✅ Foundation session caching for cost optimization
📋 Prerequisites
ZMCPTools TypeScript requires the following:
Required
- Node.js 18+ - JavaScript runtime and LanceDB native bindings
- Package Manager - npm (included), yarn, pnpm, or bun
- Claude CLI - Anthropic's Claude Code CLI
Optional
- TypeScript: For development (
npm install -g typescript
) - TSX: For development hot-reload (included in devDependencies)
Note: This TypeScript implementation includes native LanceDB vector database with no Python dependencies required.
📝 Example Commands
Here are some common workflows you can achieve with ZMCPTools:
🎨 Brand & Style Analysis
"Using the zmcp server, find the styles and generate me a brand guide called THEWEBSITE_BRAND.md"
📚 Documentation Scraping
"Scrape https://modelcontextprotocol.io/introduction -- use the selector #content-area (by ID), don't allow any subdomains, and nothing ignored (though we can ask it to ignore regex, glob, patterns to *not* get some docs), update it weekly"
🤖 Multi-Agent Orchestration
"Create a multi-agent orchestration to design and architect a ModelContextProtocol TypeScript server to do XYZ"
🔍 Project Analysis
"Analyze the project structure, and then search the .treesummary directory to see what's there"
👥 Agent Management
"Spawn an agent to do X, and then monitor its progress"
🔎 Documentation Search
"Search the documentation for X"
🌐 Browser Automation
"Create a browser session, navigate to https://example.com, take a screenshot, and analyze the page structure for accessibility issues"
🧠 Knowledge Graph Operations
"Store this implementation pattern in the knowledge graph and find related patterns we've used before"
💾 Foundation Session Optimization
"Create a multi-agent team with foundation session 'auth-refactor-2024' to refactor authentication across frontend and backend with 90% cost savings"
🔄 Development Workflow
"Start a dev server, run the test suite, and spawn an agent to fix any failing tests while monitoring progress in real-time"
📊 Cross-Agent Learning
"Analyze recent agent errors, identify patterns, and spawn a debugging agent that learns from previous failures"
🎯 Streamlined Plan System
"Create an execution plan for implementing OAuth, then execute it with coordinated agents following the plan"
The Plan system provides 4 streamlined tools for orchestration:
create_execution_plan
- Create high-level execution plans from objectivesget_execution_plan
- Retrieve plans with progress tracking via linked Tasksexecute_with_plan
- Execute plans by creating coordinated Tasks for agentslist_execution_plans
- List and monitor execution plans
Plans create Tasks for implementation - Plans are high-level orchestration templates while Tasks are specific work items assigned to agents.
📊 Data Scope
- Documentation & Websites: Shared project-wide across all repositories
- Agents, Tasks, Memory: Scoped per repository_path for isolation
- Prompts & Resources: Available globally for all projects
🎯 Multi-Agent Orchestration
Architect-Led Coordination
ZMCPTools features an AI architect that automatically analyzes objectives and spawns coordinated agent teams with proper dependencies and real-time communication.
🎯 Usage
CLI Commands
# Show help and available commands
zmcp-tools --help
# Show system status
zmcp-tools status
# Start the MCP server
zmcp-server
# Agent management
zmcp-tools agent list
zmcp-tools agent spawn -t <type> -r <repository> -d <description>
zmcp-tools agent terminate -i <agent-id>
# Task management
zmcp-tools task list
zmcp-tools task create -t <title> -d <description>
# Memory operations
zmcp-tools memory search -q <query>
zmcp-tools memory store -t <title> -c <content>
# Communication rooms
zmcp-tools room list
zmcp-tools room join -n <name>
🛠️ Development Commands
# Initial setup (one time - using pnpm)
pnpm install # Install dependencies first
pnpm run install:global # Build, link globally, and configure everything
# Alternative package managers
npm install && npm run build && npm link && zmcp-tools install
yarn install && yarn build && yarn link && zmcp-tools install
bun install && bun run build && bun link && zmcp-tools install
# Development with hot-reload
pnpm dev # Start MCP server with tsx
pnpm dev:cli # Start CLI with tsx
# Building and testing
pnpm build # Compile TypeScript to dist/
pnpm test # Run Vitest tests
pnpm test:ui # Run tests with UI
pnpm test:run # Run tests once
# Code quality
pnpm lint # ESLint checking
pnpm typecheck # TypeScript type checking
# Production
pnpm start # Start compiled MCP server
pnpm start:cli # Start compiled CLI
# Management
zmcp-tools install # Install/reinstall MCP server
zmcp-tools uninstall # Remove MCP server and settings
zmcp-tools status # Check system status
zmcp-tools help # Show all commands
# For users who installed via npx
npx zmcp-tools@latest status # Check status
npx zmcp-tools@latest uninstall # Remove installation
🌟 TypeScript Features:
🎯 Type Safety
- Full TypeScript implementation with strict mode
- Zod schemas for runtime validation
- Compile-time error checking
- IntelliSense support in IDEs
🚀 Performance
- Better-sqlite3 for high-performance database operations
- ES2022 target with modern optimizations
- Efficient memory management
- Fast development with tsx hot-reload
🧪 Testing
- Vitest for modern testing experience
- UI mode for interactive test debugging
- Coverage reports with V8 provider
- TypeScript test support out of the box
📦 Module System
- ESNext modules for tree-shaking
- Clean imports and exports
- Library mode for programmatic use
- Dual CLI and server binaries
Convenient Aliases
The following aliases are available (add to ~/.zshrc
):
alias mcp-tools="zmcp-tools"
alias mcp-server="zmcp-server"
alias mcp-status="zmcp-tools status"
alias mcp-dev="npm run dev"
⚙️ Configuration
TypeScript Configuration
# Build configuration in tsconfig.json
{
"compilerOptions": {
"target": "ES2022",
"module": "ESNext",
"moduleResolution": "bundler",
"strict": true,
"experimentalDecorators": true,
"outDir": "dist",
"rootDir": "src"
}
}
# Development scripts (works with npm/yarn/pnpm/bun)
npm run dev # Hot-reload development
npm run build # Production build
npm test # Run test suite
🎯 TypeScript Features:
- Strict Type Checking: Full type safety with strict mode enabled
- Modern ES Modules: ESNext target with bundler resolution
- Development Tools: tsx for hot-reload, Vitest for testing
- Code Quality: ESLint with TypeScript rules
- Binary Generation: Dual binaries for CLI and server
- Library Mode: Exportable as TypeScript library
Manual MCP Server Configuration (if needed)
The installer automatically configures the MCP server, but if you need to manually configure it:
# For production install (via npx)
# Server is installed at ~/.mcptools/server/index.js
# Configuration is automatic via 'claude mcp add --scope local'
# For development install
claude mcp add zmcp-tools $(pwd)/dist/server/index.js
# Verify installation
claude mcp list
# Test server directly
echo '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | node ~/.mcptools/server/index.js
🛡️ MCP Protocol Compliance
Full MCP 1.15.0 Compatibility
- JSON-RPC 2.0: Complete implementation with proper message handling
- Stdio Transport: High-performance local process communication
- Tool Definitions: 40 tools with comprehensive input schemas and validation
- Error Handling: Standardized MCP error codes and proper error propagation
- Initialization Protocol: Full handshake with capability negotiation
TypeScript MCP Implementation
Full TypeScript implementation with MCP SDK, proper error handling, and tool management.
MCP Best Practices Implemented
- Input Validation: Zod schemas for runtime type safety
- Proper Error Handling: MCP-compliant error responses with detailed messages
- Tool Annotations: Descriptive schemas with security hints where applicable
- Resource Management: SQLite database connections with proper cleanup
- Transport Security: Stdio transport for secure local communication
🏗️ Architecture
Modern CLI with TypeScript
- Commander.js Framework: Type-safe commands with automatic help generation
- Console Formatting: Colored output and structured command display
- CLI Interface: Comprehensive command structure for all operations
- Status Reporting: Real-time feedback for operations and system health
Dual-Layer Design
Layer 1: Enhanced File Operations
- Hierarchical ignore pattern system (.claudeignore > .gitignore > defaults)
- Fuzzy string matching with configurable similarity thresholds
- Cross-platform screenshot capabilities with native tool integration
Layer 2: Multi-Agent Orchestration
- Project analysis integration for intelligent code understanding
- Documentation intelligence with local vector database storage
- SQLite-based coordination with real-time communication
- Foundation Session pattern for 85-90% token cost reduction
Enhanced Project Integration
Automatic CLAUDE.md Integration:
- Unique delimiters:
<!-- zzZMCPToolszz START/END -->
- Intelligent replacement of existing sections
- Architect-led workflow guidance
- Multi-agent examples with dependencies
- Documentation-driven development patterns
Automatic Claude Hooks Integration:
- Session start context injection for instant MCP tools awareness
- Knowledge graph and core tools reminders (analyze_project_structure(), search_knowledge_graph(), plan tools)
- One-time per session to avoid context bloat
- Non-destructive settings.json merging
Per-Project Setup:
# Automatic integration during project setup
zmcp-tools install --project
# Creates/updates:
# • MCP server registration via 'claude mcp add --scope local'
# • ./.claude/commands/ (Claude commands)
# • ./.claude/hooks/ (session start context injection)
# • ./.claude/settings.json (hook configuration)
# • ./CLAUDE.md (integration guide with architect examples)
Data Storage
- Installation: Local project directory (
./dist/
) - Data Directory:
~/.mcptools/data/
(SQLite databases) - Vector Storage:
~/.mcptools/lancedb/
(LanceDB vector database) - Main Database:
~/.mcptools/data/claude_mcp_tools.db
- All Data: Agents, tasks, memory, and documentation in SQLite database
- Vector Data: Embeddings and vector indices stored in LanceDB
- Cache: Foundation session cache in memory/disk with vector index caching
🛠️ Development
# Clone and setup development environment
git clone https://github.com/zachhandley/ZMCPTools
cd ZMCPTools
pnpm install
# Quick setup
pnpm run install:global && mcp-tools install # Build, link
# Development mode
pnpm dev # Run MCP server with hot-reload
pnpm dev:cli # Run CLI with hot-reload
# Build and test
pnpm build # Compile TypeScript
pnpm test # Run test suite
# Test the binaries
node dist/index.js # MCP server
node dist/cli/index.js # CLI interface
🎨 CLI Examples
Status Display
$ zmcp-tools status
ZMCPTools Status:
✅ TypeScript Build: dist/ directory exists
✅ Data Directory: ~/.mcptools/data/
✅ SQLite Database: claude_mcp_tools.db
✅ LanceDB Vector Database: ~/.mcptools/lancedb/
✅ MCP Server: zmcp-server binary available
✅ Dependencies: @modelcontextprotocol/sdk, @lancedb/lancedb, better-sqlite3
Development Workflow
$ npm run dev
Starting TypeScript development server...
✅ TypeScript compilation successful
✅ MCP server starting on stdio
✅ SQLite databases initialized
✅ LanceDB vector database initialized
✅ Agent orchestration ready
✅ Foundation cache system active
Listening for MCP requests...
Press Ctrl+C to stop
🔍 Troubleshooting
Installation Issues
# Check prerequisites
node --version # Node.js 18+ required
pnpm --version # Package manager (or npm/yarn/bun)
claude --version # Claude CLI required
# Clean installation
rm -rf node_modules dist
pnpm install
pnpm build
# Development installation
git clone https://github.com/zachhandley/ZMCPTools
cd ZMCPTools
pnpm install && pnpm run install:global
Verification
# Check build output
ls -la dist/
node dist/index.js --help
# Check data directory
ls -la ~/.mcptools/data/
# Test MCP server
claude mcp list
echo '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | node dist/index.js
Server Connection Issues
# Test TypeScript compilation
pnpm typecheck
pnpm lint
# Test MCP server directly
node dist/index.js
# Debug with development server
pnpm dev
# Check MCP configuration
claude mcp list
claude mcp remove zmcp-tools
claude mcp add zmcp-tools $(pwd)/dist/index.js
TypeScript Issues
# Type checking errors
pnpm typecheck # Check TypeScript errors
npx tsc --noEmit --pretty # Detailed type errors
# Runtime errors
node --inspect dist/index.js # Debug with Node.js inspector
pnpm dev # Hot-reload development
# Database issues
rm -rf ~/.mcptools/data/*.db # Reset databases
node dist/index.js # Reinitialize
# Dependency issues
rm -rf node_modules pnpm-lock.yaml
pnpm install # Clean dependency install
📈 Performance & Architecture
Production Metrics
- 61 MCP Tools: Complete tool suite with full type safety and MCP 1.15.0 compliance
- Database Performance: SQLite with WAL mode and optimized connection pooling
- Vector Search: LanceDB native TypeScript bindings for <100ms semantic search
- Memory Efficiency: <75MB baseline with intelligent caching and cleanup
- Response Time: <200ms average tool execution, <50ms for cached operations
- Cost Optimization: 85-90% reduction through automatic foundation session management
Technical Architecture
TypeScript-First Design:
- Strict TypeScript with Zod schemas for runtime validation
- Modern ES modules with tree-shaking optimization
- Dual binary system (CLI + MCP server)
- Hot-reload development with tsx
Database Layer:
- SQLite with Write-Ahead Logging for performance
- Drizzle ORM for type-safe database operations
- Automatic schema migrations and connection pooling
- LanceDB vector database for semantic search
MCP Compliance:
- Full MCP 1.15.0 protocol implementation
- JSON-RPC 2.0 with proper error handling
- Stdio and HTTP transport support
- Resource and prompt management
🤝 Contributing
- Fork the repository
- Create a feature branch (
git checkout -b feature/amazing-feature
) - Make your changes with tests
- Test with Claude Code integration
- Submit a pull request
Development Guidelines
- Follow TypeScript strict mode requirements
- Add comprehensive error handling with MCP compliance
- Include tool annotations for destructive/read-only operations
- Test all changes with the actual MCP server integration
📜 License
MIT License - see LICENSE file for details.
🚀 Supercharge your Claude Code workflows with ZMCPTools - TypeScript-powered multi-agent orchestration, LanceDB vector search, type-safe development, enhanced performance, and intelligent development assistance!
28
フォロワー
65
リポジトリ
0
Gist
0
貢献数
A Model Context Protocol (MCP) server for interacting with Ghost CMS through LLM interfaces like Claude. Allow you to control your Ghost blog by simply asking Claude etc.