graph TB
    User[User] --> Client[MCP Client]
    Client --> LLM[LLM Provider]
    Client <--> Server[MCP Server]
    Server <--> Crawlab[Crawlab API]

    subgraph "MCP System"
        Client
        Server
    end

    subgraph "Crawlab System"
        Crawlab
        DB[(Database)]
        Crawlab <--> DB
    end

    class User,LLM,Crawlab,DB external;
    class Client,Server internal;

    %% Flow annotations
    LLM -.-> |Tool calls| Client
    Client -.-> |Executes tool calls| Server
    Server -.-> |API requests| Crawlab
    Crawlab -.-> |API responses| Server
    Server -.-> |Tool results| Client
    Client -.-> |Human-readable response| User

    classDef external fill:#f9f9f9,stroke:#333,stroke-width:1px;
    classDef internal fill:#d9edf7,stroke:#31708f,stroke-width:1px;

# Install from source
pip install -e .

# Or install from GitHub (when available)
# pip install git+https://github.com/crawlab-team/crawlab-mcp-server.git

# Start the MCP server
crawlab_mcp-mcp server [--spec PATH_TO_SPEC] [--host HOST] [--port PORT]

# Start the MCP client
crawlab_mcp-mcp client SERVER_URL

services:
  # ... existing Crawlab services
  
  mcp-server:
    build: ./backend/mcp-server
    ports:
      - "8000:8000"
    environment:
      - CRAWLAB_API_BASE_URL=http://backend:8000/api
      - CRAWLAB_API_TOKEN=your_api_token_here
    depends_on:
      - backend

User: "Create a new spider named 'Product Scraper' for the e-commerce project"
↓
LLM identifies intent and calls the create_spider tool
↓
MCP Server executes the API call to Crawlab
↓
Spider is created and details are returned to the user

User: "Run the 'Product Scraper' spider on all available nodes"
↓
LLM calls the run_spider tool with appropriate parameters
↓
MCP Server sends the command to Crawlab API
↓
Task is started and confirmation is returned to the user