module4

# Advanced CrewAI with MCP Sequential Thinking **Production-ready** multi-agent system for news analysis with the official MCP Sequential Thinking server. ## 🎯 What does this agent do? 1. **Parallel Search** - simultaneously searches for news from 3 sources (BBC, CNN, Reuters) 2. **MCP Sequential Thinking** - uses the official MCP server for structured analysis 3. **Synthesis of Conclusions** - generates a comprehensive report with recommendations ## 📊 Architecture ### 5-Agent System with MCP: ``` ┌─────────────────┐ │ BBC Researcher │─┐ └─────────────────┘ │ ├──► ┌──────────────────┐ ┌─────────────────┐ ┌─────────────────┐ │ │ Senior Analyst │ │ MCP Sequential │ │ CNN Researcher │─┤ │ (with MCP tools) │◄──────┤ Thinking Server │ └─────────────────┘ │ └──────────────────┘ └─────────────────┘ │ ▼ (npx @model...) ┌─────────────────┐ │ ┌──────────────────┐ │ Reuters Researcher│─┘ │ Report Synthesizer│ └─────────────────┘ └──────────────────┘ ``` ### MCP Sequential Thinking: CrewAI automatically connects to the official MCP Sequential Thinking server via `MCPServerAdapter`: ```python server_params = StdioServerParameters( command="npx", args=["-y", "@modelcontextprotocol/server-sequential-thinking"] ) with MCPServerAdapter(server_params, connect_timeout=60) as mcp_tools: analyst_agent = Agent( role='Senior News Analyst', tools=mcp_tools, # Automatically receives sequentialthinking tool ... ) ``` ## 📋 Project Structure ``` module4/ ├── README.md # Documentation ├── requirements.txt # Python dependencies (crewai-tools[mcp]!) ├── .env.example # Configuration template ├── .gitignore # Git exclusions │ ├── config.py # Configuration management ├── parallel_agent_with_mcp.py # 🎯 Main agent (5 agents + MCP) └── mcp_examples.py # 📚 Examples of various MCP servers ``` ## 🚀 Installation ### Step 1: Node.js (for MCP server) The MCP Sequential Thinking server runs via npx. Check: ```bash node --version # v18+ recommended npx --version ``` If Node.js is not installed: [https://nodejs.org/](https://nodejs.org/) ### Step 2: Python Dependencies ```bash pip install -r requirements.txt ``` **Dependencies:** - `crewai>=0.80.0` - Multi-agent framework - `crewai-tools[mcp]>=1.3.0` - MCP adapter for CrewAI - `langchain>=1.0.0` - LLM orchestration - `langchain-openai>=1.0.0` - OpenAI integration - `mcp>=1.6.0` - Model Context Protocol - `requests>=2.31.0` - HTTP requests (for Brave Search API) - `python-dotenv>=1.0.0` - Environment variables ### Step 3: Configuration ```bash cp .env.example .env ``` Edit `.env`: ```bash # Required OPENAI_API_KEY=your-openai-api-key-here BRAVE_API_KEY=your-brave-api-key-here # Obtain at https://brave.com/search/api/ # Optional DEFAULT_MODEL=gpt-4o-mini TEMPERATURE=0.7 ENABLE_MCP_THINKING=true MAX_SEARCH_RESULTS=3 ``` ## 🎮 Usage ### Basic Run ```bash python parallel_agent_with_mcp.py ``` **Expected output:** ``` 🔌 Preparing to connect to MCP Sequential Thinking server... ================================================================================ 🚀 ADVANCED CREWAI: Parallel Search + MCP Sequential Thinking ================================================================================ 📋 Creating search agents... 🔍 Analysis topic: 'artificial intelligence breakthrough' 🧠 MCP Sequential Thinking: Connecting... └─ Searching from BBC, CNN, Reuters (in parallel) ✅ MCP server connected! Available tools: 1 📋 Tools: • sequentialthinking ⚡ Starting parallel search and analysis... ``` ### MCP Sequential Thinking in Action The agent automatically uses the MCP server for a 5-step analysis: ``` Sequential Thinking MCP Server running on stdio ┌────────────────────────────────────────────────┐ │ 💭 Thought 1/5 │ ├────────────────────────────────────────────────┤ │ Identifying key themes mentioned across all │ │ sources: AI regulations, ethical concerns... │ └────────────────────────────────────────────────┘ ┌────────────────────────────────────────────────┐ │ 💭 Thought 2/5 │ ├────────────────────────────────────────────────┤ │ Finding unique insights from each source... │ └────────────────────────────────────────────────┘ ... ``` ### Programmatic Usage ```python from parallel_agent_with_mcp import run_advanced_analysis # Run analysis result = run_advanced_analysis( topic="quantum computing breakthrough" ) print(result['result']) print(f"Execution time: {result['duration']:.2f}s") print(f"MCP enabled: {result['mcp_enabled']}") ``` ## 🔧 How MCP Integration Works ### 1. MCPServerAdapter connects to npx server ```python server_params = StdioServerParameters( command="npx", args=["-y", "@modelcontextprotocol/server-sequential-thinking"], env=None ) with MCPServerAdapter(server_params, connect_timeout=60) as mcp_tools: # mcp_tools contains a list of tools from the MCP server analyst_agent = create_analyst_agent_with_mcp(mcp_tools) ``` ### 2. Agent calls MCP tool The analyst receives instructions to call `sequentialthinking`: ```python analysis_description = f''' Use the "sequentialthinking" tool for structured analysis. Step 1 - thought: "Identifying key themes..." thoughtNumber: 1, totalThoughts: 5, nextThoughtNeeded: true Step 2 - thought: "Finding unique insights..." thoughtNumber: 2, totalThoughts: 5, nextThoughtNeeded: true ... ''' ``` ### 3. MCP server visualizes the process The official MCP server automatically outputs beautiful boxes for each thought step in the console. ## 💡 Advantages of the MCP Approach ✅ **Official server** - uses @modelcontextprotocol/server-sequential-thinking ✅ **Automatic visualization** - MCP server draws thinking boxes itself ✅ **Built-in support for CrewAI** - via MCPServerAdapter ✅ **Standardized protocol** - Model Context Protocol ✅ **Easily extensible** - can add other MCP servers ## 🧪 Requirements - Python 3.10+ - Node.js 18+ (for npx) - OpenAI API Key - Internet connection (for news search and npx) ## 🌟 Other Popular MCP Servers This project demonstrates Sequential Thinking, but MCP supports many other servers! ### 📋 Official MCP Servers: #### 1. **Sequential Thinking** ⭐ (used in this project) ```bash npx -y @modelcontextprotocol/server-sequential-thinking ``` Structured step-by-step thinking for solving complex problems. #### 2. **Filesystem** ```bash npx -y @modelcontextprotocol/server-filesystem /path/to/allowed/dir ``` Safe interaction with the filesystem with access control. - Reading/writing files - Creating directories - Searching for files #### 3. **Fetch** ```bash npx -y @modelcontextprotocol/server-fetch ``` Downloading and converting web content for LLM. - Downloading HTML pages - Converting to markdown - Working with PDFs #### 4. **Git** ```bash npx -y @modelcontextprotocol/server-git --repository /path/to/repo ``` Working with Git repositories. - Reading commits - Searching history - Analyzing changes #### 5. **Memory** ```bash npx -y @modelcontextprotocol/server-memory ``` Knowledge graph memory system. - Creating entities - Relationships between concepts - Storing knowledge ### 🎯 Popular Third-Party MCP Servers: - **Google Drive** - working with Google Drive files - **Slack** - integration with Slack - **GitHub** - advanced work with GitHub - **PostgreSQL** - working with databases - **MongoDB** - NoSQL databases - **Puppeteer** - browser automation - **Brave Search** - searching via Brave - **AWS** - integration with AWS services - **Azure** - Microsoft Azure integration - **Cloudflare** - Cloudflare Workers ### 💡 Usage Examples A file `mcp_examples.py` has been created with examples of using various MCP servers: ```bash python mcp_examples.py ``` Examples include: 1. **Sequential Thinking** - structured thinking 2. **Filesystem** - working with files 3. **Fetch** - downloading web content 4. **Git** - analyzing repositories 5. **Memory** - knowledge graph ### 🔗 How to Add Another MCP Server ```python from crewai import Agent from crewai_tools import MCPServerAdapter from mcp import StdioServerParameters # Parameters for another MCP server server_params = StdioServerParameters( command="npx", args=["-y", "@modelcontextprotocol/server-NAME"] ) # Usage with MCPServerAdapter(server_params, connect_timeout=60) as mcp_tools: agent = Agent( role='Your Role', tools=mcp_tools, # Agent receives tools from MCP ... ) ``` ### 📊 Complete List of Servers: - [Official MCP Servers](https://github.com/modelcontextprotocol/servers) - [Community MCP Servers](https://github.com/punkpeye/awesome-mcp-servers) - [Marketplace](https://modelcontextprotocol.io/examples) ## 📚 Useful Links - [CrewAI MCP Documentation](https://docs.crewai.com/en/mcp/overview) - [MCP Sequential Thinking Server](https://github.com/modelcontextprotocol/servers/tree/main/src/sequentialthinking) - [Model Context Protocol](https://modelcontextprotocol.io/) - [CrewAI Documentation](https://docs.crewai.com) - [OpenAI API Reference](https://platform.openai.com/docs/api-reference) ## 🐛 Troubleshooting ### "OpenAI API key not found" **Problem:** OPENAI_API_KEY is not set **Solution:** ```bash # In .env file OPENAI_API_KEY=your-api-key-here # Or via export export OPENAI_API_KEY="your-api-key-here" ``` ### "command not found: npx" **Problem:** Node.js is not installed **Solution:** ```bash # macOS brew install node # or download from https://nodejs.org/ ``` ### "Module 'crewai_tools' not found" **Problem:** crewai-tools[mcp] is not installed **Solution:** ```bash pip install 'crewai-tools[mcp]' ``` ### Timeout when connecting to MCP **Problem:** MCP server is not starting in time **Solution:** - Check that npx works: `npx -y @modelcontextprotocol/server-sequential-thinking --version` - Increase timeout in the code: ```python with MCPServerAdapter(server_params, connect_timeout=120) as mcp_tools: ``` ## 📝 License MIT License