# llmwiki [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/lucasastorian/llmwiki  
**GitHub Stars:** 1709  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/llmwiki-2

## Description
Open Source Implementation of Karpathy's LLM Wiki. Upload documents, connect your Claude account via MCP, and have it write your wiki !

## Claude Desktop Quick Installation
No confirmed install details yet, so no config is published. Follow the setup steps in the repository above.

## Documentation

## What llmwiki does

The llmwiki MCP server turns a folder of research material into an AI-maintained wiki. It works with Markdown, PDFs, Word and PowerPoint files, spreadsheets, web pages, images, and several text and data formats. Source files remain in place, while generated pages are stored as ordinary Markdown under a `wiki/` directory.

As a member of the 🧠 Knowledge & Memory category, llmwiki is suited to personal research archives, long-lived project notes, and organizational knowledge bases. The web application provides navigation, source views, cross-links, a concept graph, and visualizations such as charts, SVGs, and Mermaid diagrams. A Chrome extension can save web pages and PDFs along with highlights and comments.

## How it works

The local CLI opens a selected workspace, starts the API and web application, and indexes the files it finds. An index and cache are stored in a hidden `.llmwiki/` directory. A background watcher can detect files added to the workspace, making uploads and dropped-in documents available to the application and MCP client.

After generating an MCP configuration for the workspace, add the printed JSON to Claude Desktop or Claude Code. The connected model can search and read the indexed material, create wiki pages, update existing pages, and maintain references between pages and their source documents. Each workspace needs its own MCP server entry.

For ongoing maintenance, schedule a Claude Routine or a local Desktop scheduled task. The routine can look for new sources, clips, and highlights, then revise affected pages. This scheduling behavior comes from the Claude environment rather than from the MCP server itself.

## Setup and configuration

Local installation requires Python 3.11 or newer and Node.js 20 or newer. Clone the repository, install the Python requirements from `api/requirements.txt` and `mcp/requirements.txt`, and install the web dependencies with npm. Run the CLI against a folder containing the material to index, for example `./llmwiki open ~/research` on macOS or Linux.

Generate client configuration with `./llmwiki mcp-config ~/research`, then place the output in the appropriate Claude Desktop or Claude Code configuration file. Local mode listens on `127.0.0.1` and does not support LAN or remote binding. A hosted mode is also available at llmwiki.app, and the project can be self-hosted.

LibreOffice is needed to extract Word and PowerPoint files. `MISTRAL_API_KEY` is optional and enables higher-quality OCR for PDF tables and complex layouts. The Chrome extension can target either the hosted application or a local workspace at `http://localhost:8000`.

## Tools and capabilities

The llmwiki MCP server supports capabilities for:

- Searching and reading indexed workspace sources.
- Creating and editing generated wiki pages.
- Preserving links between pages and their source documents.
- Exposing uploaded files, web clips, highlights, and comments to the connected model.
- Working with local document indexes through an MCP-compatible client.

Supported input includes PDF, DOCX, DOC, PPTX, PPT, XLSX, XLS, HTML, Markdown, plain text, CSV, JSON, XML, YAML, SVG, and common image formats. PDFs are extracted locally; images are stored for inline viewing and can be read by Claude when requested.

## Limitations and notes

The local API is intentionally loopback-only, so it cannot be bound for LAN access or remote connections. Office extraction depends on a local LibreOffice installation. OCR quality for difficult PDF layouts can be improved with the optional Mistral key, but that key is not required for normal operation.

LLM Wiki does not move, modify, or upload the source files in the selected folder. It adds only the generated `wiki/` directory and hidden `.llmwiki/` index and cache. The repository does not describe a fixed MCP tool-name list, so integrations should be evaluated by the documented read, search, and wiki-writing workflow rather than by assumed tool names.

_Full upstream README: https://allmcps.com/mcp/llmwiki-2/readme_