# LLM Wiki — Local Research Wiki via MCP

> LLM Wiki turns a local folder into a Markdown wiki + SQLite index, then lets Claude connect over MCP to read sources and keep pages/citations updated.

## Install

Merge the JSON below into your `.mcp.json`:

## Quick Use

1. Install and run locally:
   ```bash
   git clone https://github.com/lucasastorian/llmwiki.git
   cd llmwiki
   cd api && python -m venv .venv && source .venv/bin/activate && pip install -r requirements.txt && cd ..
   cd web && npm install && cd ..
   ./llmwiki init ~/research
   ./llmwiki serve ~/research
   ```
2. Print Claude MCP config:
   ```bash
   ./llmwiki mcp-config ~/research
   ```

## Intro

LLM Wiki turns a local folder into a Markdown wiki + SQLite index, then lets Claude connect over MCP to read sources and keep pages/citations updated.

- **Best for:** researchers who want a local, reviewable wiki that compounds over time instead of one-off summaries
- **Works with:** Python 3.11+ + Node.js 20+; local web UI; Claude via MCP; filesystem-first storage
- **Setup time:** 20–45 minutes

## Practical Notes

- Quant: the quick start calls out **Python 3.11+** and **Node.js 20+** as requirements.
- Quant: the default local ports are **:3000** (web) and **:8000** (API), with a local SQLite index under `.llmwiki/`.

## Main

A safe operating loop:

1. **Keep sources immutable**: point LLM Wiki at your research folder; it writes only to `wiki/` and `.llmwiki/`.
2. **Ask for diffs**: have Claude propose page edits and review them like code (git diff, PRs).
3. **Scope by workspace**: one folder = one MCP entry, so cross-project access stays explicit.

If you want richer PDF extraction, the README suggests optional OCR upgrades; start with the default local stack first.

### FAQ

**Q: Does it move my files?**
A: No—README says source files stay where they are; it adds `wiki/` + `.llmwiki/`.

**Q: Can I run it without cloud services?**
A: Yes—local indexing and storage are the default; basic usage needs no API keys.

**Q: How do I connect Claude?**
A: Run `./llmwiki mcp-config <folder>` and paste the JSON snippet into your Claude config.

## Source & Thanks

> Source: https://github.com/lucasastorian/llmwiki
> License: Apache-2.0
> GitHub stars: 873 · forks: 136

---

<!-- ZH -->

## 快速使用

1. 本地安装并启动：
   ```bash
   git clone https://github.com/lucasastorian/llmwiki.git
   cd llmwiki
   cd api && python -m venv .venv && source .venv/bin/activate && pip install -r requirements.txt && cd ..
   cd web && npm install && cd ..
   ./llmwiki init ~/research
   ./llmwiki serve ~/research
   ```
2. 输出 Claude MCP 配置片段：
   ```bash
   ./llmwiki mcp-config ~/research
   ```

## 简介

LLM Wiki 把本地文件夹索引成 Markdown wiki 与 SQLite 检索库，并让 Claude 通过 MCP 读取资料、生成页面并维护链接与引用；默认本地运行且不移动源文件。

- **适合谁：** 希望在本地沉淀可复核、可持续演进的研究 Wiki 的人
- **可搭配：** Python 3.11+ + Node.js 20+；本地 Web UI；Claude 通过 MCP；以文件系统为真相源
- **准备时间：** 20–45 分钟

## 实战建议

- 量化信息：Quick Start 明确要求 **Python 3.11+** 与 **Node.js 20+**。
- 量化信息：默认本地端口为 **:3000**（Web）与 **:8000**（API），并在 `.llmwiki/` 维护本地 SQLite 索引。

## 主要内容

更安全的运行方式：

1. **源文件不动**：把 LLM Wiki 指向研究目录；它只写 `wiki/` 与 `.llmwiki/`，源文件保持原样。
2. **让改动可评审**：让 Claude 先产出 wiki 编辑，再用 git diff/PR 的方式审核。
3. **按工作区隔离**：一个目录对应一个 MCP entry，跨项目访问必须显式配置。

需要更强 PDF 提取时再按 README 加 OCR；建议先用默认本地方案跑通主流程。

### FAQ

**会移动或改动我的源文件吗？**
答：不会；README 说明源文件保持不变，只新增 `wiki/` 与 `.llmwiki/`。

**不接云服务能用吗？**
答：能；默认是本地索引与存储，基础使用不需要 API key。

**怎么接入 Claude？**
答：运行 `./llmwiki mcp-config <folder>`，把输出的 JSON 片段写入 Claude 配置即可。

## 来源与感谢

> Source: https://github.com/lucasastorian/llmwiki
> License: Apache-2.0
> GitHub stars: 873 · forks: 136


---
Source: https://tokrepo.com/en/workflows/llm-wiki-local-research-wiki-via-mcp
Author: AI Open Source