casual-mcp

# Casual MCP   **Casual MCP** is a Python framework for building, evaluating, and serving LLMs with tool-calling capabilities using [Model Context Protocol (MCP)](https://modelcontextprotocol.io). ## Features - Multi-server MCP client using [FastMCP](https://github.com/jlowin/fastmcp) - OpenAI, Ollama, and Anthropic provider support (via [casual-llm](https://github.com/AlexStansfield/casual-llm)) - Recursive tool-calling chat loop - Toolsets for selective tool filtering per request - **Tool discovery** -- defer tool loading and let the LLM search for tools on demand via BM25 - Usage statistics tracking (tokens, tool calls, LLM calls) - System prompt templating with Jinja2 - CLI and API interfaces ## Installation The base package includes the core framework. Install with the extras for the LLM provider(s) you need: ```bash # Using uv uv add casual-mcp[openai] uv add casual-mcp[anthropic] uv add casual-mcp[ollama] # Multiple providers uv add casual-mcp[openai,anthropic] # All providers uv add casual-mcp[all] # Using pip pip install casual-mcp[openai] ``` For development: ```bash git clone https://github.com/casualgenius/casual-mcp.git cd casual-mcp uv sync --group dev ``` ## Quick Start 1. Create `casual_mcp_config.json`: ```json { "clients": { "openai": { "provider": "openai" } }, "models": { "gpt-4.1": { "client": "openai", "model": "gpt-4.1" } }, "servers": { "time": { "command": "python", "args": ["mcp-servers/time/server.py"] } } } ``` 2. Set your API key: `export OPENAI_API_KEY=your-key` 3. Start the server: `casual-mcp serve` 4. Make a request: ```bash curl -X POST http://localhost:8000/chat \ -H "Content-Type: application/json" \ -d '{"model": "gpt-4.1", "messages": [{"role": "user", "content": "What time is it?"}]}' ``` ## Configuration Configure clients, models, MCP servers, and toolsets in `casu…