Playing Zork has never been so boring
Build AI agents to play classic text adventure games (Zork, Colossal Cave, Enchanter, etc.) using the Model Context Protocol (MCP) and HuggingFace models.
This project provides:
- MCP Server - Exposes text adventure games as MCP tools using FastMCP
- ReAct Agent - An agent that uses MCP tools to play games with reasoning
- Submission Template - Starter code for students to implement their own solutions
- Evaluation System - Deterministic evaluation with seeded runs
- 57 Games - Zork trilogy, Infocom classics, and many more Z-machine games
Getting Started
0. Clone this space
git clone https://huggingface.co/spaces/LLM-course/Agentic-zork
This includes:
- run_agent.py: Script to run agents on text adventure games
- evaluation/: Evaluation scripts and utilities
- games/: Text adventure game environments
- submission_template/: Template for your agent submission
1. Fork the template space
Fork the template space on Hugging Face:
https://huggingface.co/spaces/LLM-course/text-adventure-template
2. Clone your fork locally
git clone https://huggingface.co/spaces/YOUR_USERNAME/text-adventure-agent
3. Implement your agent
Edit these files:
agent.py- Your ReAct agent implementation (implement theStudentAgentclass)mcp_server.py- Your MCP server implementation (add tools likeplay_action,memory, etc.)
4. Test locally
# Test MCP server interactively
fastmcp dev mcp_server.py
# Run your agent
python run_agent.py --agent . --game lostpig -v -n 20