Claude Code × SyncPen: Give Your AI a Place to Write

Claude Code is good at producing text — summaries, drafts, documentation, research notes. The catch is that the text lives in your terminal and tends to disappear when the session ends. SyncPen gives it somewhere to stay: a markdown workspace you can read from, write to, and come back to later.

Connecting the two takes a few minutes. Here is how, and why it is worth doing.

What you get

SyncPen is a minimalist, markdown-native writing tool, built for people who like to capture, organize, and connect ideas — researchers, coders, and curious minds. Once you connect it to Claude Code through the SyncPen MCP server, your terminal can:

• search your SyncPen library by title and full text
• read any document as clean markdown
• list your folders and documents
• create new documents
• update existing ones
• propose edits as tracked suggestions you accept or reject — instead of overwriting your text
• read and reply to comment threads, with @mentions
• create, rename, nest, and move folders (up to five levels deep), and move documents between them
• delete documents and folders when you are done — everything goes to trash, so nothing is lost by accident
• publish a finished draft to WordPress, Ghost, or Sanity

In short, Claude can use your own notes as context, write its results straight back into your workspace, propose changes for your approval, and keep the whole thing organized. Nothing to copy and paste, and nothing to drag around by hand.

Setup

1. Create an API key

Open SyncPen, go to Settings, then API Keys. Turn on API access and create a key. Copy it right away — you will only see it once. It starts with sp_.

2. Connect Claude Code

One line in your terminal:

claude mcp add syncpen -e SYNCPEN_API_KEY=sp_your_api_key_here -- npx -y syncpen-mcp

If you prefer to keep your configuration in a file, add this to your project .mcp.json or your global ~/.claude.json:

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

There is nothing to install. The npx -y syncpen-mcp part fetches and runs the latest package on demand.

3. Check the connection

Run /mcp in Claude Code. You should see syncpen listed, with its tools available. From there, just ask in plain language:

"List my SyncPen folders." "Find my notes about pricing." "Create a document that summarizes what we changed today."

You never name the tools yourself. You describe what you want, and Claude picks the right one.

The tools, in plain terms

Working in the same document, not just on it

The newer tools change the relationship from "the AI writes a file" to "the AI works alongside you in the document."

Instead of overwriting your text, Claude can propose edits as tracked suggestions. Your original stays put, the proposed change appears beside it, and you accept or reject each one — the same track-changes flow you would expect from a human collaborator, except the collaborator is your AI. You can ask for help and stay in control of every word that lands.

It can also take part in the conversation around the writing. Claude can read the comment threads on a draft and reply in the thread, so a question you leave in the margin — "is this number current?" — gets answered exactly where you asked it, not buried in a terminal log. When a thread is handled, it can resolve it.

And every change an agent makes is signed. The version history and each reply show whether a person or the AI wrote it, under the name of the key it used — so a shared document never leaves you guessing who did what. You stay the editor: the AI proposes, attributes its work, and waits for your call.

Why it is useful

• Your work stays put. Research and drafts do not vanish when you close the terminal. Next week, "read my notes from the kickoff" simply works.
• Research stays close. No switching tabs or hunting through folders. Ask, and the relevant document is in context.
• It proposes, it does not overwrite. Edits arrive as suggestions you approve, and comments get answered in the thread — so an AI in a shared document behaves like a careful collaborator, not a bulldozer.
• A workspace it can organize, not just fill. Claude can set up a folder structure, move documents into the right place, nest folders, and clear out what you no longer need — safely, since deletes go to trash. "Set up folders for this project and file the drafts in" becomes a single request.
• Draft to published, in one place. Claude writes the first pass; it lands in SyncPen, where you can refine it, work on it with others in real time, and publish to WordPress, Ghost, or Sanity.
• Markdown throughout. Claude thinks in markdown and SyncPen stores markdown, so nothing is lost in translation.
• Your data stays yours. The connection only touches your own account, through a key you control and can revoke at any time. And because every agent edit is signed, you can always see what it touched.

A short example

Say you owe someone a weekly summary. Try this:

"Read everything in my Reports folder, then create a document called 'Weekly Summary' with the three main themes."

Behind the scenes, Claude lists the documents in the folder, reads each one, and writes a clean summary back into SyncPen. You go from a blank page to a finished, editable document without opening a file. From there, refine a sentence or two and publish.

Prefer to keep your hand on the wheel? Ask for suggestions instead of edits:

"Read my essay draft and suggest tighter openings for each section — leave them as suggestions so I can approve them."

Claude proposes each change as a tracked suggestion. You skim them, accept the ones you like, and reject the rest — nothing changes in your document until you say so.

You can also hand it the cleanup:

"Create an 'Archive' folder, move every finished draft into it, and delete the empty folders left behind."

Claude creates the folder, moves each document, and removes the empties — and because deletes go to trash, a wrong call is easy to undo.

If something is not working

• Key rejected. Make sure API access is enabled in Settings, and that you copied the whole sp_ key with no trailing spaces. If in doubt, create a new one.
• Tools not showing. Run /mcp to check the connection, then restart Claude Code so it re-reads your configuration. Check that your JSON is valid.
• Keeping things organized. Just ask — Claude can create, rename, nest, and move folders, move documents between them, and delete what you no longer need. Deletes go to trash, so a mistaken delete is recoverable.

Where this is going

There is a larger idea underneath the setup above. Because your AI can already read, search, and write here — and because every change it makes is signed and waits for your approval — a SyncPen workspace can be more than a place to park drafts. It can be a knowledge base your agents read from and write to, and that you can actually open, read, and trust.

That last part is the difference. Most places an agent keeps what it learns are opaque — a vector store you cannot read, a memory you cannot inspect. A SyncPen workspace is the opposite: plain markdown you can open, a history that shows what the AI changed and when, and suggestions you approve rather than silent writes. "The agent wrote it, and you can see exactly what" beats "the agent wrote it, somewhere."

That is the direction we are building toward — richer retrieval, so an agent finds the right passage and not just the right document, and a workspace your agents can treat as durable memory across sessions. The throughline does not change: your AI does the work, in the open, and you stay the editor.

In short

Claude Code writes quickly. SyncPen keeps the writing — and now lets your AI suggest, comment, and publish inside it. Connect them once, and your notes, drafts, and research stay in one place: organized, attributed, and ready for you and your AI to pick up where you left off.