personal-llm-wiki-kit
Enables querying a local wiki through MCP-style tools, including search, Q\&A, and knowledge map, with a web viewer for human browsing.
README
Personal LLM Wiki Kit
Clone this repository, add one source file to raw/, run one script, and open the local viewer. The result is a small LLM Wiki that an agent can query through MCP-style tools and a human can browse in the web UI.
This repo intentionally uses a generic name and generic wording so it can be published publicly without exposing course-related search terms.
What Is Included
| Area | Path | Purpose |
|---|---|---|
| Harness | harness/RULES.md |
Operating rules for agents that maintain the wiki |
| Skill | skills/wiki-maintainer/SKILL.md |
Reusable maintenance workflow for adding new sources |
| LLM Wiki | raw/, wiki/, wiki/schema/page.schema.json |
Source folder, rendered wiki pages, and page schema |
| Visualization Tool | tools/wiki_mcp_server.py |
Local MCP-style server, wiki viewer, search, Q&A, and knowledge map |
| Ingestion Script | scripts/ingest_source.py |
Converts one text/Markdown source into a wiki page |
| Demo Evidence | demo/rendered-wiki.png |
Screenshot of a real rendered knowledge base |
Requirements
- Python 3.10 or newer
- A web browser
- No external Python package is required
30-Minute Quick Start
- Clone the repo and enter the folder.
git clone <your-public-repo-url>
cd <repo-folder>
- Put one source file into
raw/.
Example:
raw/my-first-note.md
- Convert the source into a wiki page.
python scripts/ingest_source.py raw/my-first-note.md --title "My First Note"
This creates a Markdown page in wiki/pages/.
- Start the viewer and MCP tool server.
python tools/wiki_mcp_server.py
- Open the viewer.
http://127.0.0.1:8767
- Confirm the new page appears in the left page list. Use search or the question box to test the page.
MCP Tool Endpoint
The local endpoint is:
POST http://127.0.0.1:8767/mcp
Tool List
| Tool | Description |
|---|---|
wiki.list_pages |
Returns page metadata for every wiki page |
wiki.search |
Searches title, tags, summary, and body |
wiki.get_page |
Returns one full page by slug |
wiki.related_pages |
Finds pages related by category and tags |
wiki.ask |
Produces a source-grounded answer from matching pages |
wiki.knowledge_map |
Returns nodes and links for the visualization view |
wiki.ingest_raw |
Creates a wiki page from raw text supplied by an agent |
Example: List Tools
curl -s http://127.0.0.1:8767/mcp ^
-H "Content-Type: application/json" ^
-d "{\"jsonrpc\":\"2.0\",\"id\":1,\"method\":\"tools/list\"}"
PowerShell:
Invoke-RestMethod -Uri http://127.0.0.1:8767/mcp -Method Post -ContentType "application/json" -Body '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
Example: Ask The Wiki
{
"jsonrpc": "2.0",
"id": 2,
"method": "tools/call",
"params": {
"name": "wiki.ask",
"arguments": {
"question": "What is PER and why does valuation matter?"
}
}
}
How To Add Your Own Knowledge
Use either the script or the MCP tool.
Script path:
python scripts/ingest_source.py raw/example.md --title "Example" --category note --tags example,wiki
MCP path:
{
"jsonrpc": "2.0",
"id": 3,
"method": "tools/call",
"params": {
"name": "wiki.ingest_raw",
"arguments": {
"title": "Example",
"content": "# Example\n\nImportant source notes...",
"category": "note",
"tags": ["example", "wiki"]
}
}
}
After ingestion, reload the viewer. The page is immediately available to the search, Q&A, and knowledge map.
Agent Integration Request Template
Use this prompt with an agent:
Read harness/RULES.md and skills/wiki-maintainer/SKILL.md.
Ingest raw/<filename> into the wiki.
Then verify with wiki.list_pages, wiki.search, and wiki.get_page through the MCP endpoint.
Do not invent facts that are not in the source.
Verification Checklist
python scripts/ingest_source.py raw/my-first-note.md --title "My First Note"creates a file underwiki/pages/.python tools/wiki_mcp_server.pystarts without installing packages.http://127.0.0.1:8767shows the wiki viewer.- The left page list contains the new page.
- The search box finds words from the new page.
- The knowledge map changes after adding a new page.
wiki.askreturns an answer with source page metadata.
Safety Notes
- Do not put API keys in source files or scripts.
- Do not commit private notes unless the public repository is allowed to contain them.
- The included finance sample pages are educational examples only and are not investment advice.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。