A new project called Eremoterra is flipping the script on how we interact with Large Language Models by using the Model Context Protocol (MCP) as a game engine. Rather than relying on a proprietary client, the platform allows users to plug their existing agentsβwhether that's ChatGPT, Claude, or Cursorβinto a shared, persistent world. The core concept is an 'empty Earth' where users claim parcels of land, build structures, and generate narratives through their agent's interface, effectively turning every LLM into a client for a distributed text adventure.
The MCP as Game Server Architecture
The developer describes Eremoterra as an experiment to learn MCP while having fun designing the game mechanics. The server returns a 'fact sheet' for any chosen location on the planet, which the user's agent then interprets. This architecture decouples the narrative generation from the specific model being used. Users can pick any spot on the globe, from a mountain pass to the middle of the sea, and the server provides the land, weather, and ownership status. The agent then narrates the experience, creating a personalized interface for a shared backend state.
Persistent Land Ownership and Dynamic Pricing
Unlike ephemeral chat sessions, Eremoterra introduces a persistent economy based on land claims. The world is divided into 233,280,000 parcels, each corresponding to a specific geographic area. Currently, at 0% claimed, a patch costs USD 1. The pricing model is algorithmic: the price rises by approximately 0.7% for every 0.1% of the Earth's parcels claimed, and doubles every 10% increase in total claims. This means that as the world fills up, the cost of owning a virtual plot of land could escalate to hundreds of dollars, creating a speculative economy within the text game.
Narrative Persistence and Social Features
The platform supports deep narrative persistence. Users can 'call a people' to inhabit their patch, defining their culture and beliefs, which then influences the simulation. The server maintains a chronicle of events, such as fires, deaths, and construction, which are rendered as fixed-form entries. Other players can visit these patches via shared links without needing an account, interacting with the environment and leaving 'rumors' that the owner can adopt into their official story. This creates a layered, asynchronous multiplayer experience where the world state evolves even when players are offline.
Key Takeaways
- Eremoterra uses MCP to standardize agent interaction with a persistent game server.
- Land ownership is permanent and priced dynamically based on global claim percentages.
- Users can build structures and define cultures that persist in the server's chronicle.
- The project demonstrates the viability of using agents as clients for complex, stateful applications.
The Bottom Line
Eremoterra is a clever proof-of-concept that leverages the ubiquity of AI agents to bootstrap a multiplayer world without forcing users to adopt a new app. Itβs a smart play for the MCP ecosystem.