Using AI to Read and Write Your World: MCP for Worldbuilders

Loreum includes an MCP (Model Context Protocol) server that lets AI assistants like Claude read your world data and propose changes that you review before they go live. Here is what that means and why it matters.

What is MCP?

MCP (Model Context Protocol) is an open standard that lets AI assistants connect to external tools and data sources. When you connect Loreum's MCP server to Claude, the AI can search your entities, read relationships, browse the timeline, pull your style guide, and propose new content, all through structured tools backed by your world's API.

What can you do with it?

Ask questions about your world

"Who are Luke Skywalker's allies?" Claude queries the knowledge graph and returns a natural language answer. No more scrolling through pages to find a connection.

Check for consistency

"Is there anything contradictory about the timeline of the Galactic Civil War?" Claude reads your timeline events, cross- references entity appearances, and flags potential issues.

Generate content in your style

"Write a backstory for this new character that fits with the existing faction dynamics." Claude reads the relevant factions, relationships, lore, and your style guide, then generates a backstory that is consistent with both your world and your prose style.

Create and modify world data

"Create a new location called the Iron Citadel in the Northern Wastes." Claude calls the create tool and the proposed entity lands in your review queue. You see a preview of the new entity and can accept, edit, or reject it before it touches your canon.

The Review Queue

Every write operation from the MCP server creates a pending change instead of modifying your world directly. You review these changes from a dedicated staging area in the web UI. Updates show a side-by-side diff. Creates show a full preview. Deletes show what will be removed and what references it.

Changes from a single AI session are grouped together, so you can review them in context and batch-accept if you trust the output.

How to set it up

Generate an API key from your project settings, then add the MCP server to your Claude Desktop config:

{
  "mcpServers": {
    "loreum": {
      "command": "node",
      "args": ["path/to/loreum/apps/mcp/dist/index.js"],
      "env": {
        "MCP_API_BASE_URL": "https://api.loreum.app/v1",
        "MCP_API_TOKEN": "your-api-key"
      }
    }
  }
}

Read tools (11)

Query your world data: search across all content, get individual entities with their relationships and lore, browse the storyboard, read the timeline, pull the style guide, and list lore articles. These work with both read-only and read-write API keys.

Write tools (16)

Create, update, and delete entities, relationships, lore articles, timeline events, scenes, plot points, and the style guide. All writes go through the review queue. Requires a read-write API key.

Why this matters

Most AI writing tools work in isolation. They generate text without context. With MCP, the AI has access to your entire world: characters, backstories, faction dynamics, timeline, and your writing style. The generated content is grounded in your world instead of generic.

The review queue means you stay in control. The AI proposes and you decide. No surprises in your canon.

The MCP server is free and open source. You bring your own AI and tokens. No vendor lock-in, no per-query charges from Loreum.