This MCP server enables an LLM to play classic text adventure games. The description indicates the LLM acts as the player, using exploration and problem-solving to progress through the game experience.

🛠️ Key Features

Plays classic text adventure games
LLM-controlled “player” for exploration and solving

🚀 Use Cases

Interactive text-adventure gameplay where the model drives player actions
Automated or assisted exploration and puzzle solving in classic adventure formats

⚡ Developer Benefits

Provides a straightforward integration point for text-adventure interaction
Uses an LLM as the agent to explore and solve in-game challenges

⚠️ Limitations

Available source data does not specify supported games, available tools, protocols, or configuration details

zork mcp

A text adventure interpreter as an MCP server. The AI is the player, it explores, solves puzzles, fights trolls, and collects treasure, turn by turn, through tool calls.

Games

colossal dungeon, a Zork style dungeon: a grue in the dark, a toll collecting troll, a locked grate, five treasures.
brainrot manor, a meme palace. Collect the aura artifacts. Become the main character.

Tools

list_games, available games
start_game(gameId), start/restart, returns your first room
act(command), look, go north, take the lantern, use lantern, attack troll with sword, inventory...
hint, a nudge for the current room
score, treasures, moves, win status

Run

bash

npm install && npm run build && node dist/index.js

Example session

start_game("colossal-dungeon")
act("take the lantern") → act("use lantern")
act("go north"), a troll blocks you...
act("attack troll with sword")
...find all 4 treasures to win

Add a game

Games are pure data, see src/games.ts. Rooms, items with use-effects, hints, and treasures are all declarative; no engine changes needed.

io.github.mrfentmen/zork-mcp