LLM-Wiki-MCP
Local knowledge base built from PDF lecture slides on Agentic Coding, exposed as an MCP server with search, page retrieval, and query tools.
README
LLM-Wiki_MCP
A local knowledge base built from PDF lecture slides on Agentic Coding, exposed through a browser viewer and an MCP server. No cloud services are required to run the wiki or the viewer.
What this project does
- Compiles a folder of PDFs into an interlinked Markdown wiki using
compile_wiki.pyandpypdf. - Serves the wiki as a browser-readable site via
viewer/server.py. - Exposes the wiki as an MCP server (
os_wiki_mcp_server.py) so that Claude Desktop and Claude Code can access the wiki through the MCP server.
The wiki covers nine lectures on Agentic Coding Basics: vibe coding, agentic coding, SDLC pipelines, agent specifications, subprocess calling, planning mode, orchestration, MCP, loops and hooks.
Repository structure
os-wiki-mcp/
├── os_wiki_mcp_server.py MCP server — 4 registered tools
├── compile_wiki.py PDF → Markdown compilation pipeline
├── wiki_tool.py CLI: lint, search, query
├── requirements.txt Python dependencies
├── .gitignore
│
├── viewer/
│ └── server.py Local HTTP wiki browser (port 8000)
│
├── skills/
│ ├── ingest_source.md Step-by-step ingest procedure
│ ├── lint_and_repair.md Step-by-step lint and repair procedure
│ └── query_and_writeback.md Step-by-step query and write-back procedure
│
├── wiki/
│ ├── sources/ 9 compiled source summaries
│ ├── concepts/ 8 concept pages
│ ├── entities/ 4 entity profiles
│ ├── analyses/ 2 synthesis reports
│ ├── templates/ Page scaffolds (source, concept, entity, analysis)
│ ├── index.md Master table of contents
│ └── log.md Append-only change log
│
├── raw/ 9 original PDF files (immutable)
├── docs/ Design documents
├── demo/
│ └── screenshot.png
│
├── AGENTS.md Agent persona and SOPs
├── SCHEME.md YAML frontmatter schema
├── TASK.md Task contract (Task 1 complete)
├── CLAUDE.md Environment-level rules
└── journal.md Append-only operation log
Installation
Python 3.10 or higher is required.
pip install -r requirements.txt
requirements.txt installs:
mcp>=1.0.0— MCP Python SDK (providesmcp.server.fastmcp.FastMCP)pypdf>=4.0.0— PDF text extraction used bycompile_wiki.py
The viewer also uses the markdown package for richer rendering:
pip install markdown
No Anthropic API key is needed. The current implementation uses local rule-based synthesis only.
Generating the wiki
python compile_wiki.py
This reads every PDF in raw/, extracts text with pypdf, derives concepts
and entities using keyword matching, writes Markdown pages into wiki/, and
updates wiki/index.md and wiki/log.md. Existing pages in wiki/sources/,
wiki/concepts/, and wiki/entities/ are cleared and rewritten on each run.
To verify the result:
python wiki_tool.py lint # checks for dead links and orphan pages
python wiki_tool.py search "mcp"
Running the viewer
python viewer/server.py
Open http://localhost:8000 in a browser.
Options:
--port PORT Port to listen on (default: 8000)
--host HOST Host to bind (default: 127.0.0.1)
The viewer provides:
- Sidebar listing all pages grouped by type (Sources, Concepts, Entities, Analyses)
- Client-side sidebar filter by slug text
- Page type filter buttons (All / Source / Concept / Entity / Analysis)
- Markdown rendering with fenced code blocks and tables
[[wikilink]]converted to clickable navigation links- Frontmatter shown as a metadata card at the top of each page
- Dark / light mode toggle, persisted in
localStorage - No JavaScript framework, no external CDN, no API calls
Running the MCP server
python os_wiki_mcp_server.py
The server runs as a stdio MCP process and prints its status to stderr:
[OS Wiki MCP] Server starting...
[OS Wiki MCP] Wiki root : /path/to/wiki
[OS Wiki MCP] Tools : search_wiki, get_page, list_pages, query_wiki
[OS Wiki MCP] Transport : stdio
[OS Wiki MCP] Ready.
Claude Desktop configuration
Add the following to your Claude Desktop config
(~/Library/Application Support/Claude/claude_desktop_config.json on macOS):
{
"mcpServers": {
"os-wiki": {
"command": "python",
"args": ["/absolute/path/to/os_wiki_mcp_server.py"]
}
}
}
Restart Claude Desktop. The os-wiki server appears in the tools list.
Available MCP tools
search_wiki
Searches all wiki pages (title, tags, body text) for a keyword. Returns up to 5 matching line snippets per page.
search_wiki(query="agentic")
search_wiki(query="mcp", page_type="concept")
page_type is optional and accepts: source, concept, entity,
analysis, meta.
Response fields: query, total, results (list of slug, title,
type, rel_path, snippets).
get_page
Fetches a single wiki page by its slug and returns its full content.
get_page(slug="mcp")
get_page(slug="andrej-karpathy")
Response fields: found, slug, title, type, rel_path,
frontmatter, body, outgoing_links.
If the slug does not exist, found is false and an error message
is returned.
list_pages
Lists all non-template wiki pages, optionally filtered by type.
list_pages()
list_pages(page_type="source")
Response fields: total, page_type, pages (list of slug, title,
type, rel_path). Pages are sorted by type then slug.
query_wiki
Answers a natural language question using local rule-based synthesis. Matches the question against known topic areas (vibe coding, agentic coding, MCP, harness, hooks) and returns a pre-composed cross-linked Markdown report.
query_wiki(question="What is the difference between vibe and agentic coding?")
query_wiki(question="How does MCP work?")
Response fields: question, title, slug, answer (full Markdown),
note (phase notice).
This tool does not call any external API. LLM-powered synthesis is not yet implemented.
Harness
Five files govern how agents and the compiler behave:
| File | Purpose |
|---|---|
AGENTS.md |
Defines the LLM Wiki Librarian persona, three SOPs (Ingest, Query & Write-Back, Lint), and forbidden actions |
SCHEME.md |
Canonical directory layout, slug conventions, and required YAML frontmatter fields for each page type |
TASK.md |
Project task contract and completion checklist (status: complete) |
CLAUDE.md |
Environment-level rules: CLI commands, Markdown-only constraint, harness compliance requirement |
journal.md |
Append-only log of operations; one line added after each successful task cycle |
The skills/ directory provides step-by-step checklists for the three
SOPs defined in AGENTS.md: ingesting a new source, running lint and
repairing errors, and querying with optional write-back.
Demo
This screenshot demonstrates the wiki viewer rendering the generated knowledge base.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.