wiki-mcp-server

A lightweight personal wiki Model Context Protocol (MCP) server — give Claude (or any MCP-compatible client) a place to save notes, plans, and memories as connected markdown pages with full-text search and [[wikilink]]-style backlinks.

Think of it as a tiny, file-based "second brain" your AI assistant can read and write to directly, without you needing to run a database or a hosted wiki.

Why

Most AI coding assistants either forget everything between sessions, or bolt on a flat list of memory snippets with no way to search them or see how they connect. wiki-mcp-server gives an assistant a proper (if minimal) knowledge base instead:

• Full-text search — ranked results with snippets, across titles, tags, and body text.

• Backlinks — write [[some-page-slug]] in any page and the link is tracked automatically.

• A link graph — see the whole wiki as nodes and edges, e.g. to spot orphan pages.

• Plain markdown files — every page is just a .md file with a small frontmatter header. No database, no lock-in, easy to grep, diff, or back up with git.

• Zero native dependencies — pure JavaScript, installs anywhere Node.js runs.

Related MCP server: genmem

Features

Installation

git clone https://github.com/drazisil/wiki-mcp-server.git
cd wiki-mcp-server
npm install

Requires Node.js 18+.

Usage with Claude Code / Claude Desktop

Add it to your MCP server config (e.g. ~/.claude.json for Claude Code, or the equivalent claude_desktop_config.json for Claude Desktop):

{
  "mcpServers": {
    "wiki": {
      "command": "node",
      "args": ["/absolute/path/to/wiki-mcp-server/index.js"]
    }
  }
}

Restart the client and the wiki_* tools will be available.

Configuration

How pages are stored

Each page is a single markdown file, data/<slug>.md, with a short frontmatter header:

---
title: My Page Title
tags: example, notes
created: 2026-07-16T12:00:00.000Z
updated: 2026-07-16T12:00:00.000Z
---

The page body goes here. Link to another page with [[other-page-slug]] and
it'll show up in that page's backlinks.

Because it's just files, you can version them with git, sync them with any file-sync tool, or edit them by hand — the server just re-reads the directory on every call.

Example

wiki_create_page({
  title: "Project Roadmap",
  content: "Next milestone links to [[api-redesign]].",
  tags: ["planning"]
})

wiki_search({ query: "roadmap" })
// → ranked matches with snippets

wiki_backlinks({ slug: "api-redesign" })
// → [{ slug: "project-roadmap", title: "Project Roadmap" }]

For a page too large to fetch whole, jump straight to the relevant section by content — no need to guess from heading text alone:

wiki_search_sections({ query: "budget overrun" })
// → [{ slug: "project-roadmap", sectionIndex: 2, heading: "Risks", score: 3, snippet: "...budget overrun..." }]

wiki_get_section({ slugOrTitle: "project-roadmap", index: 2 })
// → { heading: "Risks", body: "## Risks\n..." }

wiki_update_section({ slug: "project-roadmap", index: 2, content: "## Risks\nUpdated risk list.", mode: "replace" })

Or browse a page's structure directly with wiki_get_outline when you already know roughly what you're looking for:

wiki_get_outline({ slugOrTitle: "project-roadmap" })
// → { sections: [{ index: 1, level: 2, heading: "Q3 goals" }, { index: 2, level: 2, heading: "Risks" }, ...] }

License

MIT