The WorkshopNo. 03
Lorekeeper
Give it a novel, and it builds a fandom-style wiki that never spoils the chapters you haven't read.
My part: The idea, the design and the build.
- Stack
- Python
- SQLite
- Claude
- MCP
- Obsidian
- Fields
ΔΟΣ ΜΟΙ ΠΑ ΣΤΩ ΚΑΙ ΤΑΝ ΓΑΝ ΚΙΝΑΣΩ
Schematic
How it works
How Lorekeeper works, step by step:
- A book (input) to Chapters + index
- Chapters + index to Claude reads: one at a time
- Claude reads to Knowledge graph: facts + when
- Knowledge graph to Wiki (output)
- Knowledge graph to MCP for agents (output)
- Knowledge graph to SQLite (output)
Note on Knowledge graph: Only the chapters you've read
The build
Claude reads the book one chapter at a time and builds a knowledge graph of everything in it: characters, places, organisations, powers, items and lore, how they relate, and when each fact appears. From the graph come a fandom-style wiki as an Obsidian vault, an MCP server that agents can ask, with citations, and a SQLite database for anything built later.
Everything is spoiler-aware. Set your reading position, and the wiki, its articles and the agent use only the chapters up to it. Secret identities stay secret until the chapter that reveals them.
The pilot book is Lord of the Mysteries.