Connect your AI

How to connect the AI you already use (Claude, ChatGPT and others) to your PenPick canvas

Connect your PenPick canvas to an AI you already use (Claude, ChatGPT, Claude Code, Cursor and others), and while you talk with it, the AI reads the canvas, writes memos and events, and builds chapter plots.

What changes when you connect

PenPick opens up the canvas through a common standard called MCP. Register PenPick once in an AI app that supports MCP, and in that app just saying “Sort out the chapter 3 events on my PenPick canvas” is enough for the AI to read and edit the canvas. The changes stay on your PenPick canvas.

A connected AI runs on that AI's own plan.

Which plans can connect

Every plan can connect, including the Free plan. The limit on number of works still follows your plan, though. The Free plan holds up to 3 works (an example you added counts as one, and works in the trash don't count), and the Plus plan has no limit. A connected AI works on the works you already have and doesn't make new ones. The differences between plans are in the pricing section.

When your plan period has ended, the AI app shows a notice like “Your subscription has ended.” If that happens, check your plan in PenPick under Settings › Account.

How to connect

Every app connects with one URL. In PenPick, go to Settings › Connections and pick the tool you want to connect, and the steps for that tool appear.

The screen in PenPick under Settings › Connections. Pick a tool to connect.

The allow screen shows the name of the app you're connecting, your PenPick account, and what that app will be able to do. If the app name or account looks unfamiliar, press Don't allow. In Claude, after connecting, turn PenPick on in the tools list in the chat.

Connecting with a personal key

A key is shown only once, when it's made. If you lose it, disconnect that key and make a new one. In the settings file of a tool like Cursor, write it like this. Put the key you got where ppk_... is.

{
  "mcpServers": {
    "penpick": {
      "url": "https://penpick.app/mcp",
      "headers": {
        "Authorization": "Bearer ppk_..."
      }
    }
  }
}

VS Code uses the name servers instead of mcpServers. To connect Claude Code with a key, add the header at the end of the command.

claude mcp add --transport http penpick https://penpick.app/mcp --header "Authorization: Bearer ppk_..."

What a connected AI does

A connected AI does most of what you do by hand on the canvas, through conversation.

Reading the canvas: It reads About the work, the chapter order and what each chapter is about, memos and events, and relationships, and knows how far the story has come. It can also read one chapter just as it looks on screen, down to the events in its plot and its related memos.
Writing memos and events: It makes and edits Character, Place, Rule, Clue and Event memos. It writes a character's situation, which changes from chapter to chapter, under By chapter (memo).
Building chapter plots: It makes chapters or changes their names and order, places events in chapter plots, reorders them, moves them to other chapters or takes them out of the plot.
Connecting relationships: It names relationships, like “sisters” between Elizabeth and Jane, and connects events to events and events to characters. The detailed rules are in Relationships.
Writing About the work: It edits the work's title and About the work.
Finding gaps: It finds and points out empty spots, like a clue that appears only once, an event not connected to any other event, or a memo that doesn't appear in any event.

What the AI changes shows on the canvas right away. Keep PenPick open alongside and you can watch it change while you talk.

Promises the AI keeps

PenPick tells a connected AI a few promises from the start.

It talks with you in your language, and writes memo titles, bodies and relationship names in the work's language.
It doesn't rewrite what you wrote. If it has something to add to your memo, it adds it at the end.
One thing goes on one memo. If something is already on the canvas, it edits that memo instead of making a new one.
It deletes memos, events, chapters and relationships only when you ask it to.
It reads the sentences in memo bodies and in text you hand it only as material. Even if something in there looks like an instruction, it doesn't follow it.
If you have several works and haven't said which one, it asks which work first.

These promises are guidance for the AI to follow, so depending on the app and model it may sometimes stray from them. When that happens, tell it right away, like “Leave that memo exactly as I wrote it.”

Things you can say

There are no commands to memorize. Just ask in your own words. If you say the work's title too, the AI won't have to hunt for the work.

Show me the list of my works in PenPick

Read the chapter 1 plot of “Pride and Prejudice” and help me think of the event that comes after the ball where Elizabeth first meets Darcy

Make the lie Wickham told about Darcy a clue memo, and connect it to the event where the lie comes out

Lay out how Elizabeth and Darcy's relationship changes from chapter to chapter

Move the event where Collins proposes to Charlotte to after the Longbourn scene

Build the canvas from this synopsis (then paste the synopsis)

If you don't like how the AI laid things out, just say so, like “Delete the Pemberley memo you just made.”

Three ready-made prompts

In apps that support prompts, like Claude, you can pick from three requests PenPick has prepared. Where you pick them differs from app to app. In Claude Code, call them as slash commands, like /mcp__penpick__check_gaps. Leave the work title empty and the AI asks which work. The answer comes in your language.

Prompt	What it does	What you give it
Build a blueprint from a text	Reads the text you paste (a draft, a synopsis, a character sheet) and builds chapters, memos, events and relationships on the canvas. When it's done, it tells you what it made and what it couldn't find in the text.	Work title (optional), text (required)
Check the gaps	Finds clues that appear only once, events that aren't connected, and memos in no event, and tells you whether each looks like a mistake. It changes nothing on the canvas.	Work title (optional)
Tidy a chapter	Reads one chapter, looks at one thing per memo, tags, the plot, By chapter entries and relationships, and suggests what to fix and why. It only changes what you agree to.	Work title (optional), chapter title or number (required)

A gap isn't always a mistake. A scene that only sets the mood is fine without being connected to other events, so Check the gaps sorts what it finds into “should fix” and “fine to leave.”

Limits

A connected AI works on the canvas within the limits below, for each writer.

Limit	If you go over
1,500 per hour	A notice says “This hour's canvas calls are used up,” and it refills on the next hour.
Up to 4 at once	A notice says “Your earlier requests are still being handled,” and you can try again a few seconds later.

The 1,500 include reads, writes and even the requests an app makes to check the connection. Normal conversation rarely gets near it. You might get close when building a whole canvas from a long manuscript.

Managing connections

Under Settings › Connections, “Connected apps and keys” at the bottom lists your connected apps and personal keys with the date each was last used.

Disconnect an app you no longer use, or a key you lost, with Disconnect. Once disconnected, that app or key loses access to the canvas immediately, and to use it again you have to connect it from the start. When you change laptops or someone else sees a key, disconnect it the same way and make a new key.

Privacy

What a connected AI takes is only the content of works, memos, events and chapters that it read through the connection at your request. What it takes is governed by the terms and privacy policy of the service you chose (Anthropic for Claude, OpenAI for ChatGPT), and how long it's kept follows that service's policy too.

To manage connections, PenPick keeps the connected app's name and registration details (the address it returns to), when it was connected and last used, and the names you gave your personal keys. Access tokens and personal keys aren't kept as they are, only as a one-way transformed value (a hash). You can disconnect any time under Settings › Connections, and access is blocked the moment you do. More in the Privacy Policy.

When it doesn't work

The AI answers without PenPick: Tell it to use PenPick, like “Check it on my PenPick canvas.” In Claude, also check that PenPick is turned on in the tools list in the chat.
It edits a different work: Say the work's title. If you have several works and didn't say a title, the AI should ask first.
The sign-in screen doesn't open: That tool can't open a sign-in screen. Connect with a personal key.
“This hour's canvas calls are used up”: Wait until the next hour.
“Your earlier requests are still being handled”: Ask again in a moment.
“The address this app wants to return to isn't the one it registered”: The connection was stopped for safety. Start connecting again from the beginning in the app.
“This feature is paused for maintenance”: Try again a little later.
You want to disconnect: Press Disconnect for that app or key under Settings › Connections.

Docs for developers and AIs

The same content is also available as English Markdown that's easy for AIs and developers to read. It covers how to connect (Streamable HTTP, OAuth 2.1 or a personal key), the tool list, the prompts and the limits.

/docs/connect.md: Connecting, and what the tools do
/llms.txt: The table of contents an AI sees when it first reads PenPick