Tiny Agents in Python: a MCP-powered agent in ~70 lines of code

Hugging Face Blog · 1 year ago

Hugging Face extended its Python SDK to let developers build AI agents that connect to Model Context Protocol servers and discover tools in approximately 70 lines of code. The implementation requires installing `huggingface_hub[mcp]>=0.32.0` and can run agents with a single CLI command that loads configurations from the Hugging Face Hub dataset. This approach simplifies agent development by eliminating custom tool integrations—the agent's core is essentially a loop that streams LLM responses, detects tool calls, executes them via MCP sessions, and feeds results back into conversation history.