tinyagents

What is tinyagents?

TinyAgents is a minimalist implementation of LLM-powered agents that connect to Model Context Protocol (MCP) servers to access external tools. It runs on Python and is intended for developers building lightweight agent workflows.

How to use tinyagents?

Install required dependencies (Python 3.10+, mcp, huggingface-hub), set up an MCP server (e.g., the included weather server), then run either python tinytoolcallingagent.py <path_to_server_script> or python tinycodeagent.py <path_to_server_script> to start an interactive chat loop.

Key features of tinyagents

• Asynchronous operation using Python's asyncio
• Dynamic discovery and invocation of MCP server tools
• Interactive chat interface for user queries
• Integration with Hugging Face's InferenceClient (Qwen2.5-Coder-32B-Instruct)
• Two agent modes: tool-calling and code generation

Use cases of tinyagents

• Query weather alerts and forecasts via a National Weather Service MCP server
• Build custom agent workflows combining LLMs with MCP tool ecosystems
• Experiment with code-based task solving using TinyCodeAgent

FAQ from tinyagents

What models does tinyagents support?

TinyAgents uses the Qwen2.5-Coder-32B-Instruct model via Hugging Face's InferenceClient.

What MCP servers does tinyagents work with?

Both agents connect to Python or JavaScript MCP servers via stdio, enabling dynamic tool discovery from any compatible server.

Is tinyagents free to use?

—

Can tinyagents execute Python code?

The TinyCodeAgent is designed to generate Python code solutions; execution is a planned future enhancement (currently TODO).

What are the system requirements?

Python 3.10 or higher, with dependencies mcp >= 1.9.0, huggingface-hub >= 0.31.2, and optionally httpx for the weather server example.