Setup guide

Connect any LLM to this wiki

The Portable LLM Wiki is vendor-neutral. The same backend can be reached over MCP (typed tool calls for Claude Desktop, Cursor) or plain HTTP (everything else). Pick your client below.

this wiki:https://your-wiki.example.com

Open the Claude Desktop MCP config file

On macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
On Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the wiki to your mcpServers block

{
  "mcpServers": {
    "portable-llm-wiki": {
      "command": "npx",
      "args": ["-y", "portable-llm-wiki-mcp"],
      "env": {
        "WIKI_BASE_URL": "https://your-wiki.example.com"
      }
    }
  }
}

npx downloads the MCP server from npm on first run and caches it. No npm install needed.

Restart Claude Desktop

Quit and reopen Claude. You should see a small hammer icon in the input bar. Click it to confirm the wiki is connected.

Ask anything

Try: “Using the portable-llm-wiki tool, who is Avery Chen?”
The model will call query_wiki automatically. Other tools available: read_page, search_wiki, list_pages, get_neighbors.

The MCP server is a small Node.js process, published on npm as portable-llm-wiki-mcp. It speaks the Model Context Protocol to your client and HTTP to this wiki. Source: mcp/.

What you can do once connected

Query

Natural-language Q&A grounded in the wiki. Every answer cites the markdown that produced it.

> What does Avery believe about boring tools?

Browse

Read any page directly. Pages link to each other via wikilinks.

> Read [[Boring Stack First]]

Keyword search across titles + bodies + tags. Returns matches with snippets.

> Search for 'inventory'

Graph

Neighbors of any page via wikilink traversal. 1-hop or 2-hop. Useful for context expansion.

> Show neighbors of Mia Patel

Built on the Model Context Protocol for the MCP path, and plain JSON-over-HTTP for everything else. Both speak to the same FastAPI backend at /api/backend/*.

← back to the wiki