Introducing the world-models.io MCP Connector

Today, we are introducing the world-models.io MCP connector: a read-only integration that allows AI clients such as Codex, Claude Desktop, and other MCP-compatible tools to query our public world-model knowledge base.

Why MCP?

Information about world models is distributed across research papers, technical reports, model releases, laboratories, and timeline events.

The Model Context Protocol provides a standard way for AI applications to access structured external knowledge. With this connector, users can explore world-model data directly from their preferred AI client.

What the connector supports

The connector currently exposes four read-only tools:

• search_models(query, limit)
• get_model(slug)
• list_timeline_events(limit, offset)
• get_connector_status()

These tools allow clients to search the model database, retrieve detailed model information, browse historical events, and verify connector availability.

Read-only by design

The connector intentionally does not support:

• data modification;
• content publishing;
• deletion;
• moderation;
• deployment actions.

It is designed exclusively for knowledge discovery and research workflows.

Local setup

The connector runs locally using Node.js 20 or later.

Example configuration:

{
  "mcpServers": {
    "world-models-io": {
      "command": "node",
      "args": [
        "/absolute/path/to/world-models.io/mcp/server.mjs"
      ],
      "env": {
        "WORLD_MODELS_ROOT": "/absolute/path/to/world-models.io"
      }
    }
  }
}

Replace the paths with the location of your local project.

The connector reads the project’s public static indices and does not require credentials for public data access.

Example prompts

Once configured, you can ask your MCP client:

What is the latest world model timeline event?

Or:

Find world models related to autonomous driving.

You can also combine the connector with your existing research workflow to compare models, identify related developments, or explore a specific category.

Compatible clients

The same connector configuration can be used with:

• Codex;
• Claude Desktop;
• other clients supporting MCP stdio servers.

Setup instructions and ready-to-copy configuration examples are available here:

https://world-models.io/en/mcp/

Towards interoperable AI knowledge

The MCP connector is an initial step toward making world-model knowledge more accessible to both humans and AI systems.

Our goal is to provide a structured, transparent, and reusable knowledge layer for researchers, engineers, and anyone working on:

• robotics;
• autonomous systems;
• embodied AI;
• simulation;
• model-based reinforcement learning;
• generative world models.

The connector is open for exploration and currently focuses on the public data available through world-models.io.

---