Using SyncPen as Your Knowledge Base

Most knowledge bases fail the same way. Notes go in, and nothing comes out — because finding the right paragraph six months later is harder than writing it again. A knowledge base is only worth keeping if you can get back what you put in. And, increasingly, if your AI can get to it too.

SyncPen is built for both. It's a markdown writing workspace where everything you collect — articles, sources, drafts, research — lives in one place your AI can read, search, and write to directly. Here's how to use it as the knowledge base behind your work.

Get things in: capture

A knowledge base starts with capture, and capture has to be frictionless or it doesn't happen.

• The browser Clipper saves an article, a quote, or a selection straight to your Inbox.
• Email-to-document turns a forwarded newsletter, press release, or interview note into a document.
• Markdown import brings your existing notes in without a lossy conversion.

Everything lands as plain markdown — no proprietary format to fight with later.

Keep it findable: organize and search

Capture is nothing without retrieval. SyncPen gives you nested folders to structure a library, and full-text search across the body of every document, not just titles — so "that thing I read about committed-spend drawdown" surfaces the note even when the title never mentioned it.

Citations are first-class: import from Zotero or Mendeley, link sources inline, and let the bibliography build itself. For researchers and journalists, that closes the gap between collecting a source and using it.

Let your AI work in it

This is the part that changes how a knowledge base feels. SyncPen runs an MCP server, so Claude Code — or any MCP client — can operate on your library directly. Not "paste your notes into a chat." Your AI reads and writes the real documents.

Setup takes a minute. Generate an API key in Settings, then add the server to your MCP config:

{
  "mcpServers": {
    "syncpen": {
      "command": "npx",
      "args": ["-y", "syncpen-mcp"],
      "env": { "SYNCPEN_API_KEY": "sp_your_key" }
    }
  }
}

Now your AI can search your whole library by content and read any document, create and update documents, organize folders, and publish a finished piece to WordPress, Ghost, or Sanity — without leaving the workflow.

It saves tokens, not just clicks

There's a cost angle people miss. When your AI has a knowledge base it can search, you stop pasting the same documents into the chat over and over.

Without one, context is expensive. You re-paste background material every session, or the model re-reads everything just to find one answer. With SyncPen, search returns titles and IDs — a cheap index — and your AI reads only the one document it actually needs. Store once, search to find, read just-in-time.

The honest version: SyncPen doesn't compress anything. Whatever your AI reads still costs tokens. What changes is how much it has to read — selective retrieval instead of stuffing everything into the prompt. For a handful of notes that fit in a single message, it won't move the needle. For a research library you come back to across weeks, it's the difference between paying for what you need and paying for everything, every time.

The whole loop

Put together, it's one continuous motion instead of five disconnected tools:

1. Clip three sources into your Inbox.
2. Ask your AI to search the related notes and draft a briefing from them.
3. Edit it alongside the AI in the same document.
4. Publish straight to your blog.

Capture, find, write, ship — in one place, with your AI able to reach every step.

What it is, honestly

SyncPen is web-based and markdown-native. It's not a vault on your hard drive, and it's not trying to be Notion. It's a focused workspace for people who write to think — researchers, coders, and curious people who collect more than they can keep track of — and who want their AI to work inside that material, not just talk about it.

If your notes have been going in and never coming out, that's the problem worth fixing. Start with one folder, connect your AI, and see what a knowledge base can do when something can finally read all of it.