MCP Sendmail Server

# MCP SQLite Chatbot A Python chatbot that integrates with the MCP SQLite Server using local Ollama API. This chatbot can interact with SQLite databases through natural language using the Model Context Protocol (MCP). ## Features - 🤖 **Local LLM**: Uses Ollama with gpt-oss:20b model (configurable) - 🔧 **MCP Integration**: Full access to 8 SQLite CRUD tools via MCP protocol - 💬 **Interactive Chat**: Rich terminal interface with history - 🛠️ **Tool Calling**: Automatic tool detection and execution - 📝 **Smart Context**: Maintains conversation history with context management - 🎨 **Rich UI**: Beautiful terminal UI with syntax highlighting ## Architecture ``` ┌─────────────┐ ┌─────────────┐ ┌──────────────────┐ │ Chatbot │─────▶│ Ollama │ │ MCP Server │ │ (Python) │ │ API │ │ (SQLite CRUD) │ │ │◀─────│ gpt-oss │ │ │ └─────────────┘ └─────────────┘ └──────────────────┘ │ ▲ │ │ └────────────── Tool Calls ─────────────────┘ ``` For a detailed data flow diagram, see [dataflow.md](dataflow.md). ## How It Works: Key Flow Highlights **1. User Input → Chatbot** - User asks a question in natural language **2. Chatbot → Ollama Client** - Sends conversation history + available tools array via API - Tools are passed as structured data, not in prompt text **3. Ollama Client → LLM Model** - Forwards the request to the LLM (e.g., Llama 3.1) - Model is trained to recognize tools from API structure **4. LLM Analysis** - Model analyzes user intent - Decides if it needs to call tools to answer the question - Generates response with `tool_calls` field if tools needed **5. Ollama Parsing** - Ollama automatically parses LLM output into structured JSON - Extracts tool calls into `message.tool_calls` field **6. Chatbot → MCP Client** - Chatbot extracts tool name and arguments from `tool_calls` - Sends to MCP Client for execution **7. MCP Client → MCP Server** - HTTP POST to `/mcp/v1/tools/call` - Sends tool name and arguments as JSON **8. MCP Server Execution** - Executes actual business logic (SQL queries, file operations, etc.) - Interacts with SQLite database - Returns structured result **9. Result Back to Chatbot** - MCP Client returns formatted result - Chatbot appends to conversation history with `role: "tool"` **10. Loop Back to LLM** - Sends updated history (now includes tool result) back to Ollama - LLM sees the tool result in context - Generates natural language response for the user **11. Final Response to User** - No more tool calls needed - Displays formatted answer to user ### Multi-Step Tool Calling The chatbot supports **iterative tool calling** (up to 5 iterations), enabling: - Chain multiple tool calls sequentially - Use results from one tool to inform the next call - Build complex multi-step solutions automatically - Handle follow-up queries that require additional data ## Prerequisites 1. **MCP SQLite Server** running on `http://localhost:8080` - See parent directory README for setup instructions - Start server: `uvicorn src.server:app --reload --port 8080` 2. **Ollama** installed and running ```bash # Install Ollama (see https://ollama.ai) curl https://ollama.ai/install.sh | sh # Pull the model ollama pull gpt-oss:20b ``` 3. **Python 3.10+** ## Installation 1. Install dependencies: ```bash pip install -r requirements.txt ``` 2. Copy environment file (optional): ```bash cp .env.example .env ``` 3. Review and customize `config.yaml` if needed ## Configuration ### config.yaml The main configuration file controls all aspects of the chatbot: ```yaml mcp_server: url: "http://localhost:8080" # MCP server URL timeout: 30 ollama: base_url: "http://localhost:11434" # Ollama API URL model: "gpt-oss:20b" # Model to use temperature: 0.7 num_ctx: 4096 chatbot: name: "MCP Assistant" max_history: 10 # Number of message pairs to keep show_tool_calls: true # Display tool executions ``` ## Usage ### Quick Start ```bash # Terminal 1: Start MCP Server cd .. uvicorn src.server:app --reload --port 8080 # Terminal 2: Start Chatbot cd py-mcp-client python chatbot.py ``` ### Chat Commands - **General chat**: Just type your message - **quit/exit**: Exit the chatbot - **clear**: Clear conversation history - **tools**: List all available MCP tools ### Example Conversations #### Example 1: Creating a Table ``` You: Create a table called users with id, name, and email columns