AI knowledge base, defined and explained.
Short answer: a notes store organized so AI clients can read and write it directly — searchable, structured, with an authenticated API surface. Longer answer below, plus how MindWiki implements the pattern.
What it is
An AI knowledge base is your notes plus a programmatic interface AI clients can use to fetch from them. Three ingredients matter:
- Structured storage. Markdown files, a page database, or a graph database — not a pile of PDFs and screenshots. Retrieval works because the shape is predictable.
- An authenticated API.A REST surface, an MCP server, or both — with scope-bound credentials so the AI can't do more than you authorize.
- Search and links. Either keyword, vector, or both — plus first-class links between pages so the AI can follow context the way a human reader would.
Why it matters
The default failure mode of using AI for thinking work is that every conversation starts at zero. You re-paste the same project context, the same decisions, the same prior research. An AI knowledge base inverts this: your AI arrives knowing what you know.
Two practical upsides:
- Compounding context. The more you write over months, the more your AI has to ground every new answer.
- Cited answers. The AI tells you which pages it used. You can verify, edit, and link out.
How MindWiki implements it
- Vault = plain markdown files on disk plus managed cloud, keyed by file path.
- YAML frontmatter for per-page properties (title, area, tags, type, created).
[[wikilinks]]for connections between pages, with backlinks computed automatically.- A single MCP endpoint at
https://api.mindwiki.io/mcpwith OAuth consent. Claude (Pro), ChatGPT (Plus), Claude Desktop, Claude Code, and Codex connect directly. - A REST API for everything MCP exposes, with Bearer-token auth and explicit
read/writescopes for one-off scripts. - Twenty MCP tools across read, write, ask, similarity, graph, agent layer, and operating guide. See the API reference.
Worth checking when choosing a tool
- Does it expose MCP? If yes, modern AI clients can connect with OAuth and almost no configuration.
- What format does it store? Plain markdown is the most portable. Proprietary block stores lock you in.
- Are there scopes? Read-only API keys for automations should be a one-click affair.
- Does it ship a graph/backlinks view? If your AI can follow your links, your answers get sharper.
- Does the company write down what their automations do? Vague AI magic is a smell. Proposal-first automations let you stay in control.
FAQ
What is an AI knowledge base?
An AI knowledge base is a note store organized so AI clients can search, read, and update it directly — instead of starting every conversation from a blank context. The store is typically markdown or structured pages, with links between them, properties on each page, and an API or MCP endpoint the AI authenticates against.
How is it different from a regular notes app?
A regular notes app is built for human reading. An AI knowledge base adds a programmatic surface (REST API, MCP server, or both), authentication scopes for AI clients, and a structured shape (titles, links, properties) so retrieval works without scraping.
Do I need to learn anything special to use one?
No. The point is that the AI client does the lookup. You write normally; the AI fetches what it needs when you ask. The work is on the connection side, not on your daily writing.
What does MindWiki do here specifically?
MindWiki is a markdown vault on macOS and the web with a single MCP/HTTP endpoint that Claude, ChatGPT, Codex, Claude Desktop, and Claude Code connect to. Twenty MCP tools cover read, write, search, ask, similarity, graph, and proposal flows.